Multi-scene-based photovoltaic storage direct flexible household energy scheduling optimization method and system

Abstract

The application provides a multi-scene-based light storage direct flexible household energy scheduling optimization method and system, relates to the technical field of energy management, comprises obtaining energy system operation state data and electricity price, extracting energy flow and load response characteristics, performing clustering analysis to identify the current operation scene, constructing an energy balance constraint function and a cost constraint function containing a rigid constraint layer and a flexible constraint layer, and finally generating scheduling instructions in the joint scheduling space by using a multi-objective optimization solving algorithm. The application realizes intelligent and accurate scheduling of the household energy system, and improves the system operation efficiency and economy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, and particularly relates to a multi-scene-based photovoltaic storage direct flexible household energy scheduling optimization method and system. BACKGROUND

[0002] With the rapid development of distributed energy and the proposal of the concept of energy internet, household energy systems are gradually developing towards the direction of integration of diversified devices such as photovoltaic power generation, energy storage devices and intelligent loads. Such systems are usually composed of photovoltaic power generation systems, household energy storage systems, intelligent home appliance loads and energy management systems, and can realize local production, storage and consumption of energy, and have significant environmental protection and economic benefits. Under the background of power market reform and the promotion of time-of-use electricity price policy, intelligent scheduling optimization of household energy systems has become a key technology to improve the economic efficiency and reliability of the system.

[0003] Traditional household energy scheduling methods mainly adopt static optimization strategies in a single scene, and lack the ability to identify and adapt to different operating scenes. Such methods are usually based on fixed load models and preset operating parameters, and are difficult to cope with the diversity and dynamicity of actual household energy use, resulting in low system operation efficiency. SUMMARY

[0004] The multi-scene-based photovoltaic storage direct flexible household energy scheduling optimization method and system provided by the embodiments of the present application can solve the problems in the prior art.

[0005] In a first aspect, the multi-scene-based photovoltaic storage direct flexible household energy scheduling optimization method provided by the embodiments of the present application comprises:

[0006] obtaining operating state data of a household energy system and real-time electricity prices on the grid side;

[0007] extracting energy flow characteristics and load response characteristics of the operating state data, performing feature space mapping and clustering analysis, and combining historical operating scenes to identify the current operating scene of the household energy system;

[0008] According to the current operating scene, a set of scene constraint parameters is extracted, an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer is constructed, a scene cost upper limit is calculated and a cost constraint function is constructed, and a scene constraint set is obtained;

[0009] A schedulable resource domain and a schedulable time domain are constructed to obtain a joint scheduling space, a multi-objective optimization solving algorithm is used to optimize in the joint scheduling space based on the scene constraint set, and a scheduling control instruction is generated;

[0010] Based on the aforementioned scheduling and control commands, scheduling and control actions are performed on the energy storage device and flexible load, and the actual power deviation data and cost deviation data after execution are collected to form feedback information, which is then used to correct subsequent scheduling and control actions.

[0011] Extract the energy flow characteristics and load response characteristics from the operating status data, perform feature space mapping and cluster analysis, and combine them with historical operating scenarios to identify the current operating scenario of the home energy system, including:

[0012] Based on the operating status data, the charging power change rate, discharging power change rate, and power flow direction conversion frequency of each device are calculated as energy flow characteristics. The power consumption period offset and power demand fluctuation of the flexible load are calculated as load response characteristics.

[0013] The energy flow characteristics and the load response characteristics are mapped to a feature space. Multiple joint feature points are obtained through a feature fusion operator. In the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster the multiple joint feature points. Combined with preset scene recognition rules, multiple candidate scenes are obtained, and their intra-cluster density is calculated.

[0014] Obtain the labeled scene types and their corresponding joint features from the historical running scene database, calculate the feature distance between the multiple candidate scenes and each scene type in the historical running scene database, and select the candidate scene with the smallest feature space distance and the largest intra-class density as the current running scene.

[0015] In the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster the joint features. Combined with preset scene recognition rules, multiple candidate scenes are obtained, and their intra-cluster density is calculated, including:

[0016] In the feature space, the distance between each joint feature point and other joint feature points within its neighborhood is calculated to obtain the corresponding local density value, and multiple region center points and multiple region boundary points are identified.

[0017] Based on the spatial distribution relationship between the region center point and the region boundary point, multiple cluster boundaries are determined in the feature space, and multiple initial clusters are divided using the cluster boundaries as separators.

[0018] In the scene recognition rules, feature combination constraints for multiple running scenarios are defined. Statistical distribution features of all joint features within each initial cluster are extracted. When the statistical distribution features satisfy any combination constraint in the scene recognition rules, a corresponding candidate scene type identifier is matched for that initial cluster. For unmatched initial clusters, a secondary clustering decomposition is performed to obtain multiple sub-clusters that conform to the scene recognition rules, and the corresponding candidate scene type identifiers are matched to obtain multiple candidate scenes.

[0019] For each candidate scenario, the weighted average of the feature distances between all joint feature points and their corresponding centroids within the corresponding cluster is calculated to obtain the intra-cluster density of the candidate scenario.

[0020] Based on the current operating scenario, extract the scenario constraint parameter set, construct an energy balance constraint function containing rigid and flexible constraint layers, calculate the scenario cost upper limit and construct a cost constraint function, resulting in a scenario-based constraint set, including:

[0021] The adjustable range of photovoltaic power generation, the safe range of energy storage device state of charge, the response time window of flexible load, and the power limit threshold of grid interaction are extracted from the current operating scenario to obtain the scenario constraint parameter set.

[0022] Based on the scenario constraint parameter set, an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer is constructed. The rigid constraint layer constrains the sum of photovoltaic power generation and energy storage device discharge power at any time to be equal to the sum of flexible load power, energy storage device charging power and grid interaction power. The flexible constraint layer includes flexible load response delay constraint conditions and grid interaction power smoothing constraint conditions.

[0023] Based on the current operating scenario, determine the cost structure type of the cost constraint function, assign differentiated weight coefficients to the energy storage loss cost item and the electricity cost item, use the product of the expected scheduling cycle length corresponding to the current operating scenario and the unit time cost benchmark value as the upper limit of the scenario cost, and construct the cost constraint function based on the differentiated weight coefficients.

[0024] The energy balance constraint function is combined with the cost constraint function to obtain a scenario-based constraint set.

[0025] A schedulable resource domain and a schedulable time domain are constructed to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space, generating scheduling control instructions, including:

[0026] The energy output of photovoltaic power generation devices and energy storage devices in the current scheduling cycle is taken as the scheduling resource domain, and the start-stop time period of flexible load devices is taken as the scheduling time domain. The scheduling resource domain and the scheduling time domain are discretized to construct a joint scheduling spatial grid.

[0027] Based on the scenario-based constraint set, constraint verification is performed on each grid node in the joint scheduling space grid, and grid nodes that satisfy the energy balance constraint function and cost constraint function are selected to obtain the constraint feasible region.

[0028] Multiple candidate scheduling schemes are initialized within the constrained feasible region, and their electricity cost, energy storage lifetime loss value and load comfort are calculated to obtain a multi-objective optimization function. The candidate scheduling schemes are iteratively optimized, and the candidate scheduling schemes are updated based on the Pareto dominance relationship of the multi-objective optimization function during the iteration process until the convergence condition is met, and the optimal scheduling scheme is obtained.

[0029] The resource allocation information and time allocation information in the optimal scheduling scheme are converted into executable control parameters, and the adjustment range of the control parameters is set in combination with the scene transition probability to obtain the scheduling control instruction.

[0030] The candidate scheduling scheme is iteratively optimized, and during the iteration process, the candidate scheduling scheme is updated based on the Pareto dominance relationship of the multi-objective optimization function until the convergence condition is met, thereby obtaining the optimal scheduling scheme, including:

[0031] For the candidate scheduling scheme, calculate the function value of each objective dimension in its multi-objective optimization function to obtain a multi-dimensional objective vector;

[0032] In the current iteration, by comparing the multidimensional target vectors among the candidate scheduling schemes, their Pareto dominance is determined, the set of non-dominated scheduling schemes for the current iteration is obtained, the sensitivity of each non-dominated scheduling scheme is calculated, and its multidimensional target vector is updated to obtain the multidimensional target vector for the current iteration. The iteration continues until the number of non-dominated scheduling schemes is less than a preset threshold, and the Pareto optimal solution set is obtained.

[0033] Based on the historical scenario transition probability matrix, the fitness values ​​of each candidate scheduling scheme in the Pareto optimal solution set are calculated, and they are weighted and aggregated to obtain the optimal scheduling scheme.

[0034] Based on the aforementioned scheduling control commands, scheduling control actions are executed on the energy storage device and flexible load, and actual power deviation data and cost deviation data are collected after execution to form feedback information, which is used to correct subsequent scheduling control actions, including:

[0035] According to the scheduling control command, a charging and discharging control signal is sent to the power control unit of the energy storage device, a load adjustment control signal is sent to the load control unit of the flexible load, and the execution timestamps of the control actions of the energy storage device and the flexible load are recorded.

[0036] After the control action is completed, the actual charging and discharging power of the energy storage device and the actual operating power of the flexible load are collected, and the energy storage power deviation value and the load power deviation value are calculated. Time-series correlation analysis is performed on them to obtain the cumulative power deviation value. When it exceeds the cumulative threshold, the corresponding time period, deviation amplitude and deviation direction are extracted to obtain the actual power deviation data.

[0037] Based on the energy storage power deviation value and the load power deviation value, combined with the runtime deviation value, the cost deviation data is calculated. The actual power deviation data, the cost deviation data, and the control action execution timestamp are encapsulated into a data packet, which is then transmitted to the scheduling optimization module as feedback information to correct subsequent scheduling control actions.

[0038] A second aspect of the present invention provides a multi-scenario photovoltaic-storage-direct-flexible home energy dispatch optimization system, comprising:

[0039] The first unit is used to acquire operational status data of home energy systems and real-time electricity prices on the grid side;

[0040] The second unit is used to extract the energy flow characteristics and load response characteristics of the operating status data, perform feature space mapping and cluster analysis, and identify the current operating scenario of the home energy system in combination with historical operating scenarios.

[0041] The third unit is used to extract the scenario constraint parameter set according to the current operating scenario, construct an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer, calculate the scenario cost upper limit and construct a cost constraint function to obtain a scenario-based constraint set.

[0042] The fourth unit is used to construct a schedulable resource domain and a schedulable time domain to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space and generate scheduling control instructions.

[0043] The fifth unit is used to perform scheduling control actions on the energy storage device and flexible load based on the scheduling control command, and to collect actual power deviation data and cost deviation data after execution to form feedback information and correct subsequent scheduling control actions.

[0044] A third aspect of the present invention,

[0045] An electronic device is provided, comprising:

[0046] processor;

[0047] Memory used to store processor-executable instructions;

[0048] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0049] Fourth aspect of the present invention,

[0050] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0051] The beneficial effects of this application are as follows:

[0052] By employing feature space mapping and cluster analysis methods, combined with historical operating scenarios to identify the current scenario, differentiated scheduling strategies can be formulated for different operating scenarios. This improves the adaptability and accuracy of the scheduling scheme, enabling home energy systems to make more reasonable energy allocation decisions based on actual operating conditions.

[0053] An energy balance constraint function containing rigid and flexible constraint layers was constructed. The upper limit of scenario cost was calculated and a cost constraint function was constructed, so that the scheduling process satisfies both physical constraints and economy, balancing system reliability and economic benefits, and providing users with a more flexible energy management solution.

[0054] Based on the constructed schedulable resource domain and schedulable time domain to form a joint scheduling space, a multi-objective optimization algorithm is used to find the best solution in this space, realizing the coordinated optimization scheduling of photovoltaic, energy storage and load, improving the utilization rate of renewable energy, reducing the energy cost for users, and ensuring the balance of energy supply and demand and the comfort of energy use in households. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the multi-scenario photovoltaic-storage-direct-flexible home energy dispatch optimization method according to an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of the scene recognition method based on density distribution and boundary recognition. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0059] Figure 1 This is a flowchart illustrating the multi-scenario photovoltaic-storage-direct-flexible home energy dispatch optimization method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:

[0060] Obtain operational status data of home energy systems and real-time electricity prices from the grid;

[0061] The energy flow characteristics and load response characteristics of the operating status data are extracted, feature space mapping and cluster analysis are performed, and combined with historical operating scenarios, the current operating scenario of the home energy system is identified.

[0062] Based on the current operating scenario, extract the scenario constraint parameter set, construct an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer, calculate the scenario cost upper limit and construct a cost constraint function to obtain a scenario-based constraint set;

[0063] A schedulable resource domain and a schedulable time domain are constructed to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space and generate scheduling control instructions.

[0064] Based on the aforementioned scheduling and control commands, scheduling and control actions are performed on the energy storage device and flexible load, and the actual power deviation data and cost deviation data after execution are collected to form feedback information, which is then used to correct subsequent scheduling and control actions.

[0065] In one optional implementation, the energy flow characteristics and load response characteristics of the operating status data are extracted, feature space mapping and cluster analysis are performed, and combined with historical operating scenarios, the current operating scenario of the home energy system is identified, including:

[0066] Based on the operating status data, the charging power change rate, discharging power change rate, and power flow direction conversion frequency of each device are calculated as energy flow characteristics. The power consumption period offset and power demand fluctuation of the flexible load are calculated as load response characteristics.

[0067] The energy flow characteristics and the load response characteristics are mapped to a feature space. Multiple joint feature points are obtained through a feature fusion operator. In the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster the multiple joint feature points. Combined with preset scene recognition rules, multiple candidate scenes are obtained, and their intra-cluster density is calculated.

[0068] Obtain the labeled scene types and their corresponding joint features from the historical running scene database, calculate the feature distance between the multiple candidate scenes and each scene type in the historical running scene database, and select the candidate scene with the smallest feature space distance and the largest intra-class density as the current running scene.

[0069] In this specific embodiment, the home energy system includes a photovoltaic power generation device, an energy storage device, and various load devices. The system acquires the operating status data of the photovoltaic power generation device, the energy storage device, and the load devices within a preset time period, including information such as power change curves, charging and discharging status, and device switching status. Based on this operating status data, the system calculates energy flow characteristics and load response characteristics.

[0070] To extract energy flow characteristics, the charging power change rate, discharging power change rate, and power flow direction switching frequency of photovoltaic power generation devices and energy storage devices are calculated. Specifically, for the charging power change rate, taking energy storage devices as an example, the power change value is recorded within the sampling time. When the power change value is greater than a preset threshold (e.g., 0.5kW), the ratio of the power difference between two adjacent time points to the time interval is calculated as the charging power change rate. For example, if the charging power of the energy storage device increases from 2kW to 3kW within 10 seconds, the charging power change rate is 0.1kW / s. Similarly, the discharging power change rate is calculated. For the power flow direction switching frequency, the number of times the energy storage device switches from charging to discharging or from discharging to charging within a preset time period (e.g., 1 hour) is counted. For example, if the energy storage device switches from charging to discharging 5 times and from discharging to charging 4 times within one hour, the power flow direction switching frequency is 9 times.

[0071] For load response characteristics, the power consumption period offset and power demand fluctuation of flexible load devices are calculated. Flexible loads include adjustable loads such as electric vehicle charging stations and air conditioners. The power consumption period offset refers to the time difference between the actual power consumption period of the flexible load and the user's expected power consumption period. For example, if an electric vehicle is originally scheduled to charge from 18:00 to 22:00, but actually charges from 20:00 to 24:00, the power consumption period offset is 2 hours. The power demand fluctuation refers to the degree of change in power demand of the flexible load during operation, calculated as the ratio of the difference between the maximum and minimum power to the average power. For example, if an air conditioner has a maximum power of 2.5kW, a minimum power of 1.5kW, and an average power of 2kW in one hour, the power demand fluctuation is 0.5.

[0072] Mapping energy flow characteristics and load response characteristics to the feature space is achieved through a feature fusion operator. The feature fusion operator normalizes features of different dimensions to make their value ranges consistent, and then fuses them into a joint feature point through a weighted summation method. The weights can be adjusted according to the contribution of different features to scene recognition. For example, the weight of the charging and discharging power change rate is 0.3, the weight of the power flow direction conversion frequency is 0.2, the weight of the electricity consumption period offset amplitude is 0.25, and the weight of the power demand fluctuation amplitude is 0.25. In this way, the operating status data of each time period can be mapped to a joint feature point in the feature space.

[0073] In the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster multiple joint feature points. This algorithm needs to calculate the point density within a predetermined radius around each feature point and identify high-density regions as cluster centers. The cluster boundaries are determined based on the distance between points and the density gradient. For example, if the density threshold is set to 5 points per unit space and the distance threshold is 0.2 units of space, the points in the feature space can be divided into multiple clusters.

[0074] Based on pre-defined scene recognition rules, each cluster is assigned scene semantics, resulting in multiple candidate scenes. These scene recognition rules include: high charging power change rate, low power flow conversion frequency, and low electricity consumption period offset amplitude correspond to "peak-valley electricity price response scenario"; high power flow conversion frequency and high electricity consumption period offset amplitude correspond to "renewable energy consumption scenario"; and low charging and discharging power change rate and low power flow conversion frequency correspond to "energy self-sufficiency scenario," etc. For each candidate scene, its intra-cluster density is calculated, which is the reciprocal of the average distance between all points within the cluster. The smaller the distance, the higher the density, indicating a more concentrated distribution of points in the cluster and more pronounced scene characteristics.

[0075] The system retrieves labeled scenario types and their corresponding joint features from a historical operational scenario database. This database includes feature patterns for typical scenarios such as "peak-valley electricity price response," "renewable energy consumption," and "energy self-sufficiency." The system calculates the feature distance between candidate scenarios and each scenario type in the historical operational scenario database. Euclidean distance is used to calculate the distance between two points in the feature space. For example, if the joint feature point coordinates of candidate scenario A are (0.7, 0.3, 0.5, 0.2), and the feature point coordinates of the historical scenario "peak-valley electricity price response" are (0.8, 0.2, 0.6, 0.1), then the Euclidean distance between them is approximately 0.22.

[0076] The candidate scenario with the smallest feature space distance and the largest intra-class density is selected as the current operating scenario. If the distance from candidate scenario A to the historical scenario "peak-valley electricity price response" is 0.22 and the intra-class density is 8.5, and the distance from candidate scenario B to the historical scenario "renewable energy consumption" is 0.35 and the intra-class density is 6.2, then candidate scenario A is selected as the current operating scenario, that is, the current operating scenario is identified as "peak-valley electricity price response scenario".

[0077] Using the methods described above, home energy systems can accurately identify the current operating scenario, providing a basis for subsequent energy dispatching decisions and achieving efficient utilization of home energy.

[0078] Figure 2 This is a schematic diagram of a scene recognition method based on density distribution and boundary recognition. In one optional implementation, in the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster the joint features. Combined with preset scene recognition rules, multiple candidate scenes are obtained, and their intra-cluster density is calculated, including:

[0079] In the feature space, the distance between each joint feature point and other joint feature points within its neighborhood is calculated to obtain the corresponding local density value, and multiple region center points and multiple region boundary points are identified.

[0080] Based on the spatial distribution relationship between the region center point and the region boundary point, multiple cluster boundaries are determined in the feature space, and multiple initial clusters are divided using the cluster boundaries as separators.

[0081] In the scene recognition rules, feature combination constraints for multiple running scenarios are defined. Statistical distribution features of all joint features within each initial cluster are extracted. When the statistical distribution features satisfy any combination constraint in the scene recognition rules, a corresponding candidate scene type identifier is matched for that initial cluster. For unmatched initial clusters, a secondary clustering decomposition is performed to obtain multiple sub-clusters that conform to the scene recognition rules, and the corresponding candidate scene type identifiers are matched to obtain multiple candidate scenes.

[0082] For each candidate scenario, the weighted average of the feature distances between all joint feature points and their corresponding centroids within the corresponding cluster is calculated to obtain the intra-cluster density of the candidate scenario.

[0083] In this specific embodiment, for each joint feature point in the feature space, a spherical neighborhood with radius R is defined, and the Euclidean distance between the point and other joint feature points in the neighborhood is calculated. For example, when the coordinates of joint feature point P are (x1, y1, z1) and the coordinates of feature point Q in the neighborhood are (x2, y2, z2), the Euclidean distance between the two points is [(x1-x2)]. 2+(y1-y2) 2 +(z1-z2) 2 The square root of ] is used. For a feature point P, the number of points in its neighborhood whose distance is less than the threshold d is counted, and this number is defined as the local density value ρ of point P. In practical applications, R=0.5 and d=0.1 can be set to adapt to the feature distribution in different scenarios.

[0084] Identifying the region center and boundary points is achieved by analyzing the local density values ​​and distance parameters of feature points. The region center point has a high local density value and is far from other high-density points. The minimum distance from each joint feature point to points with a higher local density value is calculated and denoted as δ. When the local density value ρ of a point is greater than the density threshold ρ... t (e.g., taking 1.5 times the average density of points in the feature space) and δ is greater than the distance threshold δ t (If the average distance between points in the feature space is taken as twice the value of the average distance between points), then the point is determined to be the center point of the region. Boundary points are points located at the edge of clusters, which usually have low local density values ​​and are at a certain distance from high-density regions. When the local density value ρ of a point is less than the threshold ρ b (e.g., taking 0.5 times the average density) and δ is greater than the distance threshold δ b (If the average distance is taken as 1.5 times), then it is determined to be a boundary point.

[0085] Based on the distribution of regional center points and boundary points, the cluster boundaries are determined and initial clusters are divided. Starting from each regional center point, points in the feature space are assigned to the cluster represented by the nearest center point in descending order of density. When a boundary point is encountered, it is marked as part of the cluster boundary. By connecting these boundary points, a complete cluster boundary is formed. In a feature space containing 1000 joint feature points, 8 regional center points and 60 boundary points are identified and divided into 8 initial clusters.

[0086] A scenario recognition rule base was established, which defines feature combinations and constraints for various typical operating scenarios. For example, the constraints for the "photovoltaic self-consumption scenario" are: photovoltaic power generation greater than 1 kW and less than 5 kW, energy storage battery charging power greater than 0 kW, total load power less than 90% of photovoltaic power generation, and time period from 10:00 to 16:00. The constraints for the "nighttime energy storage discharge scenario" are: photovoltaic power generation equal to 0 kW, energy storage battery discharge power greater than 0.5 kW, total load power greater than 0.8 kW, and time period from 20:00 to 6:00 the next day. The constraints for the "high load electricity consumption scenario" are: total load power greater than 4 kW, air conditioner or electric heater load greater than 2 kW, and time period from 18:00 to 22:00.

[0087] Statistical analysis was performed on all joint feature vectors within each initial cluster to extract the distribution characteristics of each dimension, specifically the maximum, minimum, average, median, mode, and the interval of highest frequency for each dimension. Taking the first initial cluster as an example, the statistics showed that the average photovoltaic power generation was 3.6 kW, the minimum was 2.8 kW, and the maximum was 4.5 kW; the average energy storage charging power was 1.1 kW, the minimum was 0.6 kW, and the maximum was 1.8 kW; and the average total load power was 2.7 kW, the minimum was 1.9 kW, and the maximum was 3.4 kW. The frequency of these values ​​was concentrated between 12:00 and 16:00, reaching 85%.

[0088] The extracted statistical distribution features are compared one by one with the constraints of each scene in the scene recognition rule base to determine whether the statistical features of the cluster fall within the constraint range of a certain scene. For the first initial cluster, the average photovoltaic power generation of 3.6 kW is in line with the 1 to 5 kW range of the "photovoltaic self-consumption scenario", the average energy storage charging power of 1.1 kW is greater than 0 kW and meets the charging conditions, the average total load of 2.7 kW is less than 90% of the average photovoltaic power generation of 3.6 kW, i.e., 3.24 kW, and the time period is concentrated between 12:00 and 16:00, which meets the constraint of 10:00 to 16:00. Therefore, the candidate scene type identifier of "photovoltaic self-consumption scenario" is matched for this initial cluster.

[0089] In the eight initial clusters, assuming that five clusters successfully match the corresponding candidate scenario type, the statistical distribution features of the remaining three clusters do not meet the constraints of any scenario in the rule base. For these three unmatched initial clusters, a secondary clustering decomposition is required. Taking one of the unmatched clusters as an example, this cluster contains 380 joint feature vectors, and its statistical distribution shows that the photovoltaic power generation ranges from 0 kW to 3.5 kW, the energy storage status includes both charging and discharging, and the time period is unevenly distributed from morning to evening.

[0090] The unmatched cluster was further decomposed using hierarchical clustering into three sub-clusters. The first sub-cluster contained 150 feature vectors, showing an average photovoltaic power generation of 2.8 kW, an average energy storage charging power of 0.9 kW, and a time period concentrated between 11:00 and 14:00. This sub-cluster met the constraint of "PV self-consumption scenario" and was matched with the corresponding candidate scenario type identifier. The second sub-cluster contained 120 feature vectors, showing an average photovoltaic power generation of 0.2 kW, an average energy storage discharge power of 1.3 kW, and a time period concentrated between 17:00 and 19:00. This sub-cluster met the constraint of "evening energy storage discharge scenario" and was matched with the corresponding candidate scenario type identifier. The third sub-cluster contained 110 feature vectors, showing a photovoltaic power generation close to 0 kW, energy storage in standby mode, a low average total load of 0.6 kW, and a time period concentrated between 6:00 and 8:00 in the morning. This sub-cluster met the constraint of "low load standby scenario" and was matched with the corresponding candidate scenario type identifier.

[0091] After matching or decomposing all initial clusters, 11 candidate scenarios were identified. For each candidate scenario, the intra-cluster tightness was calculated to evaluate the quality of the clustering. The calculation method is to traverse each joint feature vector within the cluster corresponding to the candidate scenario and calculate the Euclidean distance between the vector and the cluster centroid. For example, if the photovoltaic power generation of a certain feature vector is 3.4 kW and the photovoltaic power generation of the centroid is 3.8 kW, the difference in this dimension is 0.4 kW. The square of the difference is calculated for all six dimensions, and the square root of the six squared values ​​is taken to obtain the Euclidean distance from the feature vector to the centroid as 0.62.

[0092] The distance values ​​calculated from all feature vectors within a cluster are weighted and averaged. The weights for the photovoltaic power generation dimension are set to 0.25, energy storage status dimension to 0.25, total load dimension to 0.2, ambient temperature dimension to 0.15, and time period dimension to 0.15. The distance value of each feature vector is multiplied by its corresponding dimension weight, and the sum of these products is divided by the total number of feature vectors to obtain the intra-cluster density value for the candidate scene. For example, if a candidate scene contains 420 feature vectors and the calculated weighted average distance is 0.58, then the intra-cluster density of this candidate scene is 0.58. A lower intra-cluster density value indicates a more concentrated distribution of feature vectors within the scene, and thus a higher reliability for scene recognition.

[0093] In one optional implementation, based on the current operating scenario, a set of scenario constraint parameters is extracted, an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer is constructed, the upper limit of scenario cost is calculated, and a cost constraint function is constructed to obtain a scenario-based constraint set, including:

[0094] The adjustable range of photovoltaic power generation, the safe range of energy storage device state of charge, the response time window of flexible load, and the power limit threshold of grid interaction are extracted from the current operating scenario to obtain the scenario constraint parameter set.

[0095] Based on the scenario constraint parameter set, an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer is constructed. The rigid constraint layer constrains the sum of photovoltaic power generation and energy storage device discharge power at any time to be equal to the sum of flexible load power, energy storage device charging power and grid interaction power. The flexible constraint layer includes flexible load response delay constraint conditions and grid interaction power smoothing constraint conditions.

[0096] Based on the current operating scenario, determine the cost structure type of the cost constraint function, assign differentiated weight coefficients to the energy storage loss cost item and the electricity cost item, use the product of the expected scheduling cycle length corresponding to the current operating scenario and the unit time cost benchmark value as the upper limit of the scenario cost, and construct the cost constraint function based on the differentiated weight coefficients.

[0097] The energy balance constraint function is combined with the cost constraint function to obtain a scenario-based constraint set.

[0098] When extracting constraint parameter sets from the current operating scenario, it is necessary to obtain the real-time status of the photovoltaic power generation system. Data collected by the photovoltaic array monitoring module is used to determine the adjustable range of photovoltaic power generation. For example, under sunny summer weather conditions, if the rated power of the photovoltaic system is 200kW, considering equipment efficiency and environmental factors, the actual adjustable range is 0-180kW. Simultaneously, the state of charge (SOC) data of the energy storage device is collected through the battery management system (BMS) to determine the safe operating range. For lithium iron phosphate battery packs, the safe SOC range is typically set at 15%-85% to ensure battery life. For flexible loads, the response time window is determined based on the equipment operating status collected by the load controller and the user-set priority. For example, an air conditioning system is allowed to adjust its power by ±20% within 30 minutes in summer, and an electric vehicle charging pile can flexibly adjust its charging power within 60 minutes. The grid interaction power limit threshold is determined based on the grid connection agreement signed with the grid company and the capacity of local power facilities. For example, the upper limit of grid interaction power is 250kW, and the lower limit is -150kW (negative values ​​indicate power fed into the grid).

[0099] Based on the extracted scenario constraint parameter set, an energy balance constraint function is constructed, comprising a rigid constraint layer and a flexible constraint layer. The rigid constraint layer ensures energy balance at any scheduling time point t. Specifically, a constraint condition is established requiring that the sum of photovoltaic power generation and energy storage device discharge power must equal the sum of flexible load power, energy storage device charging power, and grid interaction power. This constraint guarantees energy conservation within the microgrid, avoiding energy waste or supply-demand imbalance. Taking a microgrid with a scheduling interval of 15 minutes as an example, at a certain moment, the photovoltaic output is 110kW, the energy storage discharge power is 30kW, the flexible load power consumption is 80kW, and the energy storage charging power is... If the power is 25kW, then the grid interaction power should be 35kW to ensure energy balance. The flexible constraint layer includes two types of constraints: the flexible load response delay constraint controls the time characteristics of load regulation. For example, after the air conditioning load is issued a regulation command, the actual response must be completed within a preset 30-minute time window, allowing for gradual adjustment of power according to user comfort settings. The grid interaction power smoothing constraint limits the rate of change of grid interaction power in adjacent time periods. For example, it stipulates that within an adjacent 15-minute dispatch interval, the change in grid interaction power should not exceed 20% of the rated interaction capacity, i.e., 50kW, in order to reduce the impact on the grid and improve system stability.

[0100] Based on the current operating scenario, the structural type of the cost constraint function is further determined and weighting coefficients are set. According to the microgrid's operating objectives, the cost structure can be divided into economic-oriented or renewable energy consumption-oriented types. For the economic-oriented type, weighting coefficients are assigned to the energy storage loss cost item and the electricity consumption cost item, respectively. For example, during peak electricity price periods (such as 14:00-17:00 in the afternoon), the weight of energy storage loss cost is set to 0.3 and the weight of electricity consumption cost is 0.7; during off-peak electricity price periods (such as 01:00-05:00 in the morning), the weight is adjusted to 0.6 for energy storage loss cost and 0.4 for electricity consumption cost, in order to encourage energy storage charging and reduce grid electricity consumption. By analyzing historical dispatch data and current electricity pricing policies, a benchmark value for the unit time cost within the expected dispatch cycle (e.g., 24 hours) is calculated, for example, set at 120 yuan / hour. This benchmark value is multiplied by the expected dispatch cycle of 24 hours to obtain the upper limit of the scenario cost of 2880 yuan, which serves as the boundary condition for cost constraints. Based on differentiated weighting coefficients, a cost constraint function is constructed, which includes the energy storage cycle loss cost and the grid purchase cost under time-of-use pricing. For a 100kWh energy storage system, the cost of each complete charge-discharge cycle is approximately 20 yuan; while the grid electricity price is set according to time periods, such as 1.2 yuan / kWh during peak hours, 0.8 yuan / kWh during normal hours, and 0.4 yuan / kWh during off-peak hours.

[0101] By combining energy balance constraints with cost constraints, a complete scenario-based constraint set is formed. This constraint set satisfies the energy balance requirements of the physical system while also considering economy and flexibility, providing a solution boundary for subsequent optimization algorithms. The constraint set is stored in the microgrid energy management system database in the form of a data structure for use by the optimization scheduling module. This scenario-based constraint construction method significantly improves the scheduling flexibility and economy of the microgrid, enabling the system to adaptively adjust its operating strategy according to different operating scenarios.

[0102] In one optional implementation, a schedulable resource domain and a schedulable time domain are constructed to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space to generate scheduling control instructions, including:

[0103] The energy output of photovoltaic power generation devices and energy storage devices in the current scheduling cycle is taken as the scheduling resource domain, and the start-stop time period of flexible load devices is taken as the scheduling time domain. The scheduling resource domain and the scheduling time domain are discretized to construct a joint scheduling spatial grid.

[0104] Based on the scenario-based constraint set, constraint verification is performed on each grid node in the joint scheduling space grid, and grid nodes that satisfy the energy balance constraint function and cost constraint function are selected to obtain the constraint feasible region.

[0105] Multiple candidate scheduling schemes are initialized within the constrained feasible region, and their electricity cost, energy storage lifetime loss value and load comfort are calculated to obtain a multi-objective optimization function. The candidate scheduling schemes are iteratively optimized, and the candidate scheduling schemes are updated based on the Pareto dominance relationship of the multi-objective optimization function during the iteration process until the convergence condition is met, and the optimal scheduling scheme is obtained.

[0106] The resource allocation information and time allocation information in the optimal scheduling scheme are converted into executable control parameters, and the adjustment range of the control parameters is set in combination with the scene transition probability to obtain the scheduling control instruction.

[0107] In this embodiment, the dispatchable resource domain includes photovoltaic power generation devices and energy storage devices. The energy output that these devices can provide during the current dispatch cycle constitutes the dispatchable resource domain. For example, during a 24-hour dispatch cycle, the photovoltaic power generation device can generate an hourly expected power generation sequence [0, 0, 0, 0, 0, 0.5, 2.3, 4.7, 6.8, 7.9, 8.2, 8.0, 7.5, 6.2, 4.3, 2.1, 0.6, 0, 0, 0, 0, 0, 0, 0] kWh based on weather forecasts and historical data. The energy storage device has a capacity of 10 kWh, a charge / discharge efficiency of 0.95, an initial charge of 5 kWh, and a maximum charge / discharge power of 3 kW. These parameters together define the energy boundary of the dispatchable resource.

[0108] The schedulable time domain consists of the start-stop periods of flexible load devices. For example, a washing machine needs to run once between 6:00 and 22:00, with a running time of 2 hours and a power of 1.2 kW; an air conditioner needs to maintain the room temperature within a set range between 13:00 and 20:00, with an adjustable power range of 0.8 to 2.5 kW; and an electric vehicle needs to be charged to more than 80% between 18:00 and 7:00 the next day, with a charging power of 3.5 kW and a required charging time of 4 hours.

[0109] When discretizing the schedulable resource domain and time domain, the 24-hour scheduling cycle is divided into 48 time slots, each lasting 30 minutes. The charging and discharging power of the energy storage device is discretized in 0.5 kW increments, resulting in power options of [-3, -2.5, -2, ..., 0, ..., 2.5, 3] kW, where negative values ​​represent discharging and positive values ​​represent charging. The start-up and shutdown times of flexible loads are discretized in time slots. For example, the startable time slots for a washing machine are [12, 13, ..., 44], corresponding to 6:00 to 22:00. Through this discretization process, a multi-dimensional grid-like joint scheduling space is constructed.

[0110] The scenario-based constraint set includes energy balance constraints and cost constraints. The energy balance constraint requires that within each time slot, the sum of photovoltaic power generation and energy storage discharge minus energy storage charging must equal the sum of fixed load and flexible load. For example, in a specific time slot, if photovoltaic power generation is 5 kWh, fixed load is 2 kWh, and flexible load is 1.5 kWh, then the energy storage charging / discharging amount needs to be adjusted to 1.5 kWh of charging to ensure energy balance. The cost constraint considers the grid's time-of-use pricing to avoid excessive electricity consumption during peak periods. For example, if the electricity price is [0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.8, 0.8, 1.2, 1.2, 1.2, 1.2, 1.2, 0.8, 0.8, 0.8, 0.8, 0.8, 1.2, 1.2, 1.2, 0.8, 0.8, 0.8, 1.2, 1.2, 1.2, 0.8, 0.8], If the price is 0.5 yuan / kWh, the dispatching scheme should minimize the amount of electricity purchased by the grid during the high-price periods of 9:00-12:00 and 18:00-21:00.

[0111] During the constraint verification process, each grid node is checked one by one to see if it meets the constraints. The energy storage capacity constraint is verified to ensure that the energy storage capacity is not less than 1 kWh and not more than 10 kWh at any time. The load demand constraint is verified, such as the washing machine must complete a full run within a specified time window and the air conditioner must maintain the room temperature within a set range. Through these verifications, grid nodes that meet all constraints are selected to form the constraint feasible region.

[0112] When initializing multiple candidate scheduling schemes within the constrained feasible region, a random initialization strategy is used to generate 20 different scheduling schemes. Each scheme includes the charging and discharging plan of the energy storage device and the start-up and shutdown time arrangement of the flexible load. For example, an initial scheduling scheme arranges for the washing machine to start at 14:00, the air conditioner to operate at a power of [1.2, 1.5, 1.8, 2.0, 1.5, 1.2, 0.9] kW from 13:00 to 20:00, and the electric vehicle to charge from 22:00 to 2:00 the next day.

[0113] For each candidate solution, three optimization target values ​​are calculated: electricity cost, energy storage life loss, and load comfort. Electricity cost is calculated by summing the grid purchase volume of each time slot and multiplying it by the corresponding electricity price. Energy storage life loss is evaluated based on the charging and discharging depth and frequency. For example, the loss of an 80% deep discharge is twice that of a 40% deep discharge. Load comfort is calculated based on the deviation between the actual operating time of the flexible load and the user's expected time. The smaller the deviation, the higher the comfort.

[0114] A multi-objective optimization algorithm is used for iterative optimization. In each iteration, candidate scheduling schemes are updated based on Pareto dominance. If scheme A is not inferior to scheme B on all objectives and is superior to B on at least one objective, then A dominates B. By retaining non-dominated schemes and eliminating dominated schemes, the algorithm gradually approaches the Pareto optimal solution set. The iterative process continues until the convergence condition is met, such as the optimal solution set changing by no more than 1% in 10 consecutive iterations, or reaching the maximum number of iterations of 100.

[0115] The optimal scheduling scheme is selected from the Pareto optimal solution set to balance the objectives. For example, the scheme schedules the washing machine to start at 16:00 to take advantage of the surplus photovoltaic power and the lower electricity price period; it schedules the electric vehicle to start charging at 23:00 to take advantage of the off-peak electricity price at night; and the energy storage device charges when the photovoltaic power generation is sufficient and discharges during the peak electricity price period, which reduces electricity costs and reduces dependence on the grid.

[0116] When converting the optimal scheduling scheme into control parameters, in order to cope with the uncertainties in actual operation, an adjustment range is set in combination with the scenario transition probability. For example, if the actual output of photovoltaic power generation deviates from the predicted value due to weather changes, the energy storage charging and discharging power is set to an adjustment range of ±10%, and the flexible load start-up and shutdown time is set to an adjustment range of ±1 time slot. Finally, a control instruction set containing device ID, operation time, operation type, parameter value and its adjustment range is generated and sent to each energy device for execution to realize intelligent energy scheduling and control.

[0117] In one optional implementation, the candidate scheduling scheme is iteratively optimized. During the iteration process, the candidate scheduling scheme is updated based on the Pareto dominance of the multi-objective optimization function until the convergence condition is met, thereby obtaining the optimal scheduling scheme, including:

[0118] For the candidate scheduling scheme, calculate the function value of each objective dimension in its multi-objective optimization function to obtain a multi-dimensional objective vector;

[0119] In the current iteration, by comparing the multidimensional target vectors among the candidate scheduling schemes, their Pareto dominance is determined, the set of non-dominated scheduling schemes for the current iteration is obtained, the sensitivity of each non-dominated scheduling scheme is calculated, and its multidimensional target vector is updated to obtain the multidimensional target vector for the current iteration. The iteration continues until the number of non-dominated scheduling schemes is less than a preset threshold, and the Pareto optimal solution set is obtained.

[0120] Based on the historical scenario transition probability matrix, the fitness values ​​of each candidate scheduling scheme in the Pareto optimal solution set are calculated, and they are weighted and aggregated to obtain the optimal scheduling scheme.

[0121] In this embodiment, the function values ​​of each objective dimension in the multi-objective optimization function of the candidate scheduling scheme are calculated to obtain a multi-dimensional objective vector. The objective dimensions include minimizing electricity costs, minimizing energy storage battery lifespan loss, and maximizing load comfort. Taking a candidate scheduling scheme as an example, this scheme stipulates that in the next 24 hours, photovoltaic power generation will prioritize supplying household loads, and excess electricity will be used to charge energy storage batteries. The energy storage batteries will discharge during peak electricity price periods, and any shortfall will be purchased from the grid.

[0122] Calculate the objective function value of the electricity cost for this candidate dispatch scheme. Calculate the cumulative value of the electricity purchased from the grid and the corresponding electricity price for each time period within 24 hours. Assuming that the scheme purchases 15 kWh of electricity from the grid during the peak period (6 hours) at a unit price of 0.8 yuan per kWh, 22 kWh during the normal period (10 hours) at a unit price of 0.5 yuan per kWh, and 18 kWh during the valley period (8 hours) at a unit price of 0.3 yuan per kWh, then the objective function value of the electricity cost is 15 multiplied by 0.8 plus 22 multiplied by 0.5 plus 18 multiplied by 0.3, which equals 28.4 yuan.

[0123] The objective function value for the lifespan loss of the energy storage battery is calculated. The degree of loss is evaluated based on the number of charge-discharge cycles, depth of charge-discharge, and charge-discharge power. In this candidate scheme, the energy storage battery performs 3 charging operations with charging depths of 60%, 45%, and 30% of the state of charge, respectively, and 2 discharging operations with discharging depths of 70% and 55%, respectively. The depth of charge-discharge is multiplied by the corresponding loss coefficient and then summed. The loss coefficient is 0.02 for every 10% of the charging depth and 0.025 for every 10% of the discharging depth. The total loss value is calculated as follows: 0.6 x 6 x 0.02 + 0.45 x 4.5 x 0.02 + 0.3 x 3 x 0.02 + 0.7 x 7 x 0.025 + 0.55 x 5.5 x 0.025 = 0.487.

[0124] The objective function value for load comfort is calculated, and the load comfort is quantified based on the deviation between the actual operating power and the expected operating power of various electrical loads. In this scheme, the air conditioner load operates at its rated power of 2.5 kW for 8 hours, at a reduced power of 1.8 kW for 12 hours, and is in a shut-off state for 4 hours within 24 hours, while the expected operating state is to operate at the rated power of 2.5 kW all day. The water heater load operates at its rated power of 2.0 kW for 6 hours and is in a standby state of 0.05 kW for 18 hours within 24 hours, while the expected operating state is to operate at its rated power for 8 hours. The washing machine load operates at its rated power of 1.2 kW for a predetermined 2 hours, which fully meets the expected operating state. Converting air conditioner load power deviation into comfort loss, a loss of 0.4 points is calculated for every 0.1 kW deviation during reduced power operation, and 3 points per hour when off. The calculated comfort loss for air conditioner load is 12 hours multiplied by 0.7 kW divided by 0.1 multiplied by 0.4, plus 4 hours multiplied by 3, equaling 45.6 points. Converting water heater load power deviation into comfort loss, a loss of 2 points per hour is calculated for every hour not operating as expected, resulting in a comfort loss of 4 points for 2 hours multiplied by 2. Washing machine load fully meets expectations with no loss, resulting in a total load comfort loss of 49.6 points. This translates to a load comfort objective function value of 100 minus 49.6, equaling 50.4.

[0125] The function values ​​of the three objective dimensions are combined to form a multi-dimensional objective vector for the candidate scheduling scheme, which is recorded as electricity cost of 28.4 yuan, energy storage loss of 0.487, and load comfort score of 50.4 points. Fifty candidate scheduling schemes are generated, each scheme corresponding to a multi-dimensional objective vector. In the current iteration, we compare the Pareto dominance relationships among these 50 schemes. The determination method is that for any two schemes, if scheme A is not inferior to scheme B in all objective dimensions, and is superior to scheme B in at least one objective dimension, then scheme A is said to dominate scheme B. Taking the multidimensional objective vector of scheme A as electricity cost of 26.5 yuan, energy storage loss of 0.512, and load comfort score of 68.3 points, and the multidimensional objective vector of scheme B as electricity cost of 28.4 yuan, energy storage loss of 0.487, and load comfort score of 50.4 points as an example, scheme A is better because its electricity cost of 26.5 yuan is less than that of scheme B (28.4 yuan) and its load comfort score of 68.3 points is greater than that of 50.4 points. However, its energy storage loss of 0.512 is worse than that of 0.487. Therefore, scheme A does not dominate scheme B.

[0126] By iterating through all candidate schemes and comparing each pairwise, scheduling schemes not dominated by any other scheme are selected, forming the set of non-dominated scheduling schemes for the current iteration. Assuming the current iteration yields 22 non-dominated scheduling schemes, the sensitivity of each scheme is calculated, measuring the influence of its position in the target space on neighboring solutions. This is done by finding the five closest non-dominated schemes and calculating the average distance between the scheme and these five schemes across various target dimensions. For example, the distances between a certain non-dominated scheme and its five closest schemes in the electricity cost dimension are 1.2 yuan, 1.5 yuan, 0.9 yuan, 1.8 yuan, and 1.3 yuan, with an average distance of 1.34 yuan. In the energy storage loss dimension, the distances are 0.032, 0.048, 0.055, 0.041, and 0.039, with an average distance of 0.043. In the load comfort dimension, the distances are 4.2 points, 5.6 points, 3.8 points, 6.1 points, and 4.5 points, with an average distance of 4.84 points.

[0127] The average distances across the three dimensions are weighted and summed according to their respective normalized weights: electricity cost (0.4), energy storage loss (0.3), and load comfort (0.3). The sensitivity of the proposed solution is then calculated. Specifically, the average distance for electricity cost (1.34 yuan) is divided by the maximum range of 50 yuan for electricity cost, yielding a normalized value of 0.0268. The average distance for energy storage loss (0.043) is divided by the maximum range of 1.0 for energy storage loss, yielding a normalized value of 0.043. The average distance for load comfort (4.84 points) is divided by the maximum range of 100 points for load comfort, yielding a normalized value of 0.0484. These three normalized values ​​are then multiplied by their respective weights and summed to obtain the sensitivity: 0.0268 × 0.4 + 0.043 × 0.3 + 0.0484 × 0.3 = 0.0397.

[0128] Based on the sensitivity values ​​of each non-dominated scheduling scheme, its multidimensional objective vector is updated and adjusted. Schemes with higher sensitivity indicate that they are in a relatively isolated position in the objective space and need to enhance their objective optimization efforts. For schemes with a sensitivity greater than 0.03, their electricity cost objective value is optimized by multiplying it by an adjustment coefficient of 0.96, their load comfort objective value is improved by multiplying it by an adjustment coefficient of 1.02, and their energy storage loss objective value is improved by multiplying it by an adjustment coefficient of 0.98. After the update, a new multidimensional objective vector for the current iteration is obtained. For example, the original electricity cost of a certain scheme is updated from 26.5 yuan to 25.44 yuan, the load comfort score is updated from 68.3 points to 69.67 points, and the energy storage loss is updated from 0.512 to 0.502.

[0129] Using the updated multidimensional objective vector as input for the next iteration, the Pareto dominance of candidate scheduling schemes is re-evaluated, and a new set of non-dominated scheduling schemes is selected. After 5 iterations, the number of non-dominated scheduling schemes decreases from the initial 22 to 15, then to 9 in the 8th iteration, 6 in the 11th iteration, and 4 in the 14th iteration. When the number of non-dominated scheduling schemes is less than a preset threshold of 5, the iteration process is terminated, and the current 4 non-dominated scheduling schemes are taken as the Pareto optimal solution set.

[0130] A scenario transition probability matrix is ​​constructed based on historical operational data to record the transition probabilities between different operational scenarios. This matrix contains 11 identified scenarios, and each element in the matrix represents the probability of transitioning from one scenario to another. For example, the probability of transitioning from a photovoltaic self-consumption scenario to an evening energy storage discharge scenario is 0.68, the probability of transitioning from a nighttime energy storage discharge scenario to a low-load early morning scenario is 0.82, and the probability of transitioning from a high-load electricity consumption scenario to a photovoltaic self-consumption scenario is 0.53.

[0131] Fitness values ​​were calculated for the four candidate scheduling schemes in the Pareto optimal solution set. The fitness value reflects the robustness and applicability of the scheduling scheme in future scenario transitions. The calculation method is to apply the scheduling scheme to the historical scenario transition sequence and statistically analyze the comprehensive objective performance of the scheme under each scenario transition path. Taking scheme C as an example, it was applied to the scenario transition records of the past 30 days, and the three-dimensional objective function value of the scheme was calculated after each scenario transition. The weighted average was taken for all transition cases. Assuming that under the path of transitioning from photovoltaic self-consumption scenario to evening energy storage discharge scenario, the electricity cost of scheme C is 24.8 yuan, the energy storage loss is 0.435, and the load comfort score is 72.5 points. Under the path of transitioning from nighttime energy storage discharge to early morning low load scenario, the electricity cost is 19.6 yuan, the energy storage loss is 0.398, and the load comfort score is 78.2 points. The weighting is combined with the scenario transition probabilities of 0.68 and 0.82.

[0132] The suitability score for this scheme is 24.8 x 0.68 + 19.6 x 0.82 = 32.936 yuan for electricity cost, 0.435 x 0.68 + 0.398 x 0.82 = 0.622 for energy storage loss, and 72.5 x 0.68 + 78.2 x 0.82 = 113.424 points for load comfort. These three dimensions are then weighted according to a set value: electricity cost 0.4, energy storage... Weighted aggregation of loss (0.3) and load comfort (0.3) yields the following results: electricity cost adaptability score normalized to 32.936 divided by 50 equals 0.659; energy storage loss adaptability score normalized to 0.622 divided by 1.0 equals 0.622; load comfort adaptability score normalized to 113.424 divided by 200 equals 0.567; and the overall adaptability value is 0.659 multiplied by 0.4, plus 0.622 multiplied by 0.3, plus 0.567 multiplied by 0.3, equals 0.620.

[0133] The fitness values ​​of the four schemes in the Pareto optimal solution set were calculated to be 0.620, 0.698, 0.605 and 0.672, respectively. The scheme with the highest fitness value, i.e., the scheme with a fitness value of 0.698, was selected as the optimal scheduling scheme and output to the home energy management system for execution. This scheme has the best overall performance in future scenario transitions, and can effectively reduce electricity costs and reduce the lifespan of energy storage batteries while ensuring load comfort.

[0134] In one optional implementation, based on the scheduling control command, scheduling control actions are performed on the energy storage device and flexible load, and actual power deviation data and cost deviation data after execution are collected to form feedback information, which is used to correct subsequent scheduling control actions, including:

[0135] According to the scheduling control command, a charging and discharging control signal is sent to the power control unit of the energy storage device, a load adjustment control signal is sent to the load control unit of the flexible load, and the execution timestamps of the control actions of the energy storage device and the flexible load are recorded.

[0136] After the control action is completed, the actual charging and discharging power of the energy storage device and the actual operating power of the flexible load are collected, and the energy storage power deviation value and the load power deviation value are calculated. Time-series correlation analysis is performed on them to obtain the cumulative power deviation value. When it exceeds the cumulative threshold, the corresponding time period, deviation amplitude and deviation direction are extracted to obtain the actual power deviation data.

[0137] Based on the energy storage power deviation value and the load power deviation value, combined with the runtime deviation value, the cost deviation data is calculated. The actual power deviation data, the cost deviation data, and the control action execution timestamp are encapsulated into a data packet, which is then transmitted to the scheduling optimization module as feedback information to correct subsequent scheduling control actions.

[0138] In this specific embodiment, scheduling control actions are performed on the energy storage device and flexible load according to the scheduling control command. After receiving the scheduling control command, a charging and discharging control signal is sent to the power control unit of the energy storage device. For example, when the system requires the energy storage device to discharge at a power of 2MW during the period from 14:00 to 16:00, the scheduling system will generate a control signal containing information such as the discharge period and power value, and send it to the power control unit of the energy storage device. At the same time, the system will also send a load adjustment control signal to the load control unit of the flexible load, such as instructing a certain electric vehicle charging pile to reduce the charging power from the originally planned 100kW to 60kW during the period from 19:00 to 21:00. While sending the control signal, the system will automatically record the execution timestamp of the control action, including the command issuance time, command reception confirmation time, and command start execution time, forming timestamp data in the format of "2023-07-15 13:58:45", which is used for subsequent deviation analysis and time series correlation.

[0139] After the control action is completed, the actual charging and discharging power of the energy storage device and the actual operating power of the flexible load are acquired in real time through the acquisition device. For example, for the above 2MW discharge command, the actual output power of the energy storage device is acquired once per minute. The acquired data are 1.92MW, 1.95MW, 1.88MW, etc. The actual power is compared with the command power to calculate the power deviation value. Taking the above data as an example, the power deviation values ​​are -0.08MW, -0.05MW, and -0.12MW, respectively. For the flexible load, its actual operating power is also acquired and the deviation value is calculated. For example, if the actual operating power is 62kW, 64kW, and 61kW, the corresponding deviation values ​​are 2kW, 4kW, and 1kW.

[0140] A time-series correlation analysis is performed on these power deviation values ​​to calculate the cumulative power deviation. For example, for the power deviation of the energy storage device, the deviation values ​​are accumulated in time sequence to obtain the cumulative deviation curve. When the cumulative deviation exceeds a preset threshold, the time period is marked as a significant deviation interval. Assuming the preset cumulative threshold is 0.5MWh, when the cumulative power deviation of the energy storage device reaches 0.52MWh in the 30 minutes from 15:00 to 15:30, the time period (15:00-15:30), the deviation amplitude (0.52MWh), and the deviation direction (negative deviation, i.e., the actual discharge is less than the planned value) will be extracted as the actual power deviation data.

[0141] Based on the acquired power deviation value, cost deviation data is calculated. For energy storage devices, the cost deviation is calculated based on the power deviation and operating time deviation, combined with electricity price information. For example, if the actual power of the energy storage device during the discharge phase is lower than the planned value, resulting in a shortfall of 0.52 MWh in electricity sales, the cost deviation is 390 yuan based on the then-current electricity price of 0.75 yuan / kWh. For flexible loads, such as an electric vehicle charging power 4 kW higher than the planned value for 2 hours, resulting in an additional 8 kWh of electricity consumption, the cost deviation is 4.8 yuan based on an electricity cost of 0.6 yuan / kWh.

[0142] The aforementioned actual power deviation data, cost deviation data, and control action execution timestamps are integrated into a single data packet. This data packet structure includes: a controlled object identifier (e.g., "Energy Storage Device-01", "Charging Pile-A12"), a summary of the control instruction content, an execution timestamp, an actual power data sequence, a power deviation value sequence, a cumulative deviation value, a significant deviation period, and a cost deviation value. The system transmits this data packet as feedback information to the scheduling optimization module.

[0143] After receiving feedback information, the scheduling optimization module corrects subsequent scheduling control actions. Specifically, the module analyzes the temporal distribution characteristics of power deviation data and equipment response characteristics, and adjusts the equipment response model parameters. For example, for the persistent negative deviation of energy storage devices between 15:00 and 15:30, the system analysis indicates that the actual dischargeable power decreases due to increased energy storage temperature. Therefore, in subsequent scheduling plans, the upper limit of available energy storage power during this period is adjusted to 1.85MW instead of the rated 2MW. For the positive deviation of flexible loads, it is found to be a fixed-mode deviation caused by user behavior. Therefore, a 5% margin is added in subsequent scheduling, adjusting the target value from 60kW to 57kW to offset the impact of users' habitual over-consumption of electricity.

[0144] Through the aforementioned feedback correction mechanism, the scheduling strategy can be continuously optimized, improving the scheduling accuracy and economy of energy storage devices and flexible loads.

[0145] This invention provides a multi-scenario photovoltaic-storage-direct-flexible home energy dispatch optimization system, comprising:

[0146] The first unit is used to acquire operational status data of home energy systems and real-time electricity prices on the grid side;

[0147] The second unit is used to extract the energy flow characteristics and load response characteristics of the operating status data, perform feature space mapping and cluster analysis, and identify the current operating scenario of the home energy system in combination with historical operating scenarios.

[0148] The third unit is used to extract the scenario constraint parameter set according to the current operating scenario, construct an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer, calculate the scenario cost upper limit and construct a cost constraint function to obtain a scenario-based constraint set.

[0149] The fourth unit is used to construct a schedulable resource domain and a schedulable time domain to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space and generate scheduling control instructions.

[0150] The fifth unit is used to perform scheduling control actions on the energy storage device and flexible load based on the scheduling control command, and to collect actual power deviation data and cost deviation data after execution to form feedback information and correct subsequent scheduling control actions.

[0151] A third aspect of the present invention provides an electronic device, comprising:

[0152] processor;

[0153] Memory used to store processor-executable instructions;

[0154] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0155] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0156] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-scenario photovoltaic-storage-direct-drive-flexible home energy dispatch optimization method, characterized in that, include: Obtain operational status data of home energy systems and real-time electricity prices from the grid; Extract the energy flow characteristics and load response characteristics from the operating status data, perform feature space mapping and cluster analysis, and combine them with historical operating scenarios to identify the current operating scenario of the home energy system, including: Based on the operating status data, the charging power change rate, discharging power change rate, and power flow direction conversion frequency of each device are calculated as energy flow characteristics. The power consumption period offset and power demand fluctuation of the flexible load are calculated as load response characteristics. The energy flow characteristics and the load response characteristics are mapped to a feature space. Multiple joint feature points are obtained through a feature fusion operator. In the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster the multiple joint feature points. Combined with preset scene recognition rules, multiple candidate scenes are obtained, and their intra-cluster density is calculated. Obtain the labeled scene types and their corresponding joint features from the historical running scene database, calculate the feature distance between the multiple candidate scenes and each scene type in the historical running scene database, and select the candidate scene with the smallest feature space distance and the largest intra-class density as the current running scene. Based on the current operating scenario, extract the scenario constraint parameter set, construct an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer, calculate the scenario cost upper limit and construct a cost constraint function to obtain a scenario-based constraint set; A schedulable resource domain and a schedulable time domain are constructed to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space and generate scheduling control instructions. Based on the aforementioned scheduling and control commands, scheduling and control actions are performed on the energy storage device and flexible load, and the actual power deviation data and cost deviation data after execution are collected to form feedback information, which is then used to correct subsequent scheduling and control actions. In the feature space, a clustering algorithm based on density distribution and boundary recognition is used to cluster the joint features. Combined with preset scene recognition rules, multiple candidate scenes are obtained, and their intra-cluster density is calculated, including: In the feature space, the distance between each joint feature point and other joint feature points within its neighborhood is calculated to obtain the corresponding local density value, and multiple region center points and multiple region boundary points are identified. Based on the spatial distribution relationship between the region center point and the region boundary point, multiple cluster boundaries are determined in the feature space, and multiple initial clusters are divided using the cluster boundaries as separators. In the scene recognition rules, feature combination constraints for multiple running scenarios are defined. Statistical distribution features of all joint features within each initial cluster are extracted. When the statistical distribution features satisfy any combination constraint in the scene recognition rules, a corresponding candidate scene type identifier is matched for that initial cluster. For unmatched initial clusters, a secondary clustering decomposition is performed to obtain multiple sub-clusters that conform to the scene recognition rules, and the corresponding candidate scene type identifiers are matched to obtain multiple candidate scenes. For each candidate scenario, the weighted average of the feature distances between all joint feature points and their corresponding centroids within the corresponding cluster is calculated to obtain the intra-cluster density of the candidate scenario.

2. The method according to claim 1, characterized in that, Based on the current operating scenario, extract the scenario constraint parameter set, construct an energy balance constraint function containing rigid and flexible constraint layers, calculate the scenario cost upper limit and construct a cost constraint function, resulting in a scenario-based constraint set, including: The adjustable range of photovoltaic power generation, the safe range of energy storage device state of charge, the response time window of flexible load, and the power limit threshold of grid interaction are extracted from the current operating scenario to obtain the scenario constraint parameter set. Based on the scenario constraint parameter set, an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer is constructed. The rigid constraint layer constrains the sum of photovoltaic power generation and energy storage device discharge power at any time to be equal to the sum of flexible load power, energy storage device charging power and grid interaction power. The flexible constraint layer includes flexible load response delay constraint conditions and grid interaction power smoothing constraint conditions. Based on the current operating scenario, determine the cost structure type of the cost constraint function, assign differentiated weight coefficients to the energy storage loss cost item and the electricity cost item, use the product of the expected scheduling cycle length corresponding to the current operating scenario and the unit time cost benchmark value as the upper limit of the scenario cost, and construct the cost constraint function based on the differentiated weight coefficients. The energy balance constraint function is combined with the cost constraint function to obtain a scenario-based constraint set.

3. The method according to claim 1, characterized in that, A schedulable resource domain and a schedulable time domain are constructed to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space, generating scheduling control instructions, including: The energy output of photovoltaic power generation devices and energy storage devices in the current scheduling cycle is taken as the scheduling resource domain, and the start-stop time period of flexible load devices is taken as the scheduling time domain. The scheduling resource domain and the scheduling time domain are discretized to construct a joint scheduling spatial grid. Based on the scenario-based constraint set, constraint verification is performed on each grid node in the joint scheduling space grid, and grid nodes that satisfy the energy balance constraint function and cost constraint function are selected to obtain the constraint feasible region. Multiple candidate scheduling schemes are initialized within the constrained feasible region, and their electricity cost, energy storage lifetime loss value and load comfort are calculated to obtain a multi-objective optimization function. The candidate scheduling schemes are iteratively optimized, and the candidate scheduling schemes are updated based on the Pareto dominance relationship of the multi-objective optimization function during the iteration process until the convergence condition is met, and the optimal scheduling scheme is obtained. The resource allocation information and time allocation information in the optimal scheduling scheme are converted into executable control parameters, and the adjustment range of the control parameters is set in combination with the scene transition probability to obtain the scheduling control instruction.

4. The method according to claim 3, characterized in that, The candidate scheduling scheme is iteratively optimized, and during the iteration process, the candidate scheduling scheme is updated based on the Pareto dominance relationship of the multi-objective optimization function until the convergence condition is met, thereby obtaining the optimal scheduling scheme, including: For the candidate scheduling scheme, calculate the function value of each objective dimension in its multi-objective optimization function to obtain a multi-dimensional objective vector; In the current iteration, by comparing the multidimensional target vectors among the candidate scheduling schemes, their Pareto dominance is determined, the set of non-dominated scheduling schemes for the current iteration is obtained, the sensitivity of each non-dominated scheduling scheme is calculated, and its multidimensional target vector is updated to obtain the multidimensional target vector for the current iteration. The iteration continues until the number of non-dominated scheduling schemes is less than a preset threshold, and the Pareto optimal solution set is obtained. Based on the historical scenario transition probability matrix, the fitness values ​​of each candidate scheduling scheme in the Pareto optimal solution set are calculated, and they are weighted and aggregated to obtain the optimal scheduling scheme.

5. The method according to claim 1, characterized in that, Based on the aforementioned scheduling control commands, scheduling control actions are executed on the energy storage device and flexible load, and actual power deviation data and cost deviation data are collected after execution to form feedback information, which is used to correct subsequent scheduling control actions, including: According to the scheduling control command, a charging and discharging control signal is sent to the power control unit of the energy storage device, a load adjustment control signal is sent to the load control unit of the flexible load, and the execution timestamps of the control actions of the energy storage device and the flexible load are recorded. After the control action is completed, the actual charging and discharging power of the energy storage device and the actual operating power of the flexible load are collected, and the energy storage power deviation value and the load power deviation value are calculated. Time-series correlation analysis is performed on them to obtain the cumulative power deviation value. When it exceeds the cumulative threshold, the corresponding time period, deviation amplitude and deviation direction are extracted to obtain the actual power deviation data. Based on the energy storage power deviation value and the load power deviation value, combined with the runtime deviation value, the cost deviation data is calculated. The actual power deviation data, the cost deviation data, and the control action execution timestamp are encapsulated into a data packet, which is then transmitted to the scheduling optimization module as feedback information to correct subsequent scheduling control actions.

6. A multi-scenario photovoltaic-storage-direct-drive-flexible home energy dispatching optimization system, used to implement the method as described in any one of claims 1-5, characterized in that, include: The first unit is used to acquire operational status data of home energy systems and real-time electricity prices on the grid side; The second unit is used to extract the energy flow characteristics and load response characteristics of the operating status data, perform feature space mapping and cluster analysis, and identify the current operating scenario of the home energy system in combination with historical operating scenarios. The third unit is used to extract the scenario constraint parameter set according to the current operating scenario, construct an energy balance constraint function containing a rigid constraint layer and a flexible constraint layer, calculate the scenario cost upper limit and construct a cost constraint function to obtain a scenario-based constraint set. The fourth unit is used to construct a schedulable resource domain and a schedulable time domain to obtain a joint scheduling space. Based on the scenario-based constraint set, a multi-objective optimization algorithm is used to find the optimal solution within the joint scheduling space and generate scheduling control instructions. The fifth unit is used to perform scheduling control actions on the energy storage device and flexible load based on the scheduling control command, and to collect actual power deviation data and cost deviation data after execution to form feedback information and correct subsequent scheduling control actions.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images