A micro-grid power prediction method and system based on aggregation state constraint

Abstract

A micro-grid power prediction method and system based on aggregation state constraint, comprising: obtaining micro-grid multi-source operation data to construct an aggregation state vector; performing multi-scale decomposition on the historical power sequence of the micro-grid grid-connected point, and rearranging different scale components according to the aggregation state vector, fusing the rearranged scale components and the aggregation state vector to obtain state-driven input samples; establishing a prediction network; inputting the input samples into the prediction network to obtain the power prediction value of the target period; according to the aggregation state vector, constructing a corresponding constraint attribute set for the power prediction value, taking the power prediction value of the target period and the constraint attribute set as the joint prediction output, and extending to the target prediction time domain. The present application can improve the consistency and applicability of micro-grid power prediction under different aggregation operation states, and provide support for micro-grid dispatching and operation optimization.

Description

Technical Field

[0001] This invention belongs to the field of microgrid dispatching technology, and more specifically, relates to a microgrid power prediction method and system based on aggregation state constraints. Background Technology

[0002] With the large-scale integration of distributed photovoltaic, wind power, electric vehicles, and energy storage resources into the power distribution network, the output and load characteristics of the system exhibit significant multi-source coupling and multi-scale fluctuation features. The spatiotemporal evolution of distributed energy resource (DER) aggregation is highly nonlinear, nonstationary, and stochastic, making it difficult for traditional centralized prediction models to accurately reflect the dynamic correlations and mutual influences among multiple resources.

[0003] Existing microgrid power forecasting methods typically construct input samples directly based on historical power sequences or by simply piecing together external information such as weather and load. These methods fail to effectively characterize the impact of distributed resource aggregation operation status on the sample structure and output target. For microgrid scenarios with significant changes in aggregation status, using existing unified sample representation methods easily leads to sample distribution drift. This means that the distribution of power data in actual power forecasting differs from the distribution of historical power data used to train the forecasting model, resulting in decreased forecasting model performance or even complete failure, thus affecting the consistency and usability of the forecast results.

[0004] In addition, existing methods often only output the power point value for the target time period, which is difficult to reflect the corresponding operating mode, fluctuation level or resource feasibility boundary of the prediction results in a synchronous manner, thus limiting the direct application of the prediction results in microgrid scheduling and operation control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a microgrid power prediction method and system based on aggregation state constraints, which improves the consistency and applicability of microgrid power prediction under different aggregation operating states, and provides support for microgrid scheduling and operation optimization.

[0006] The present invention adopts the following technical solution.

[0007] A first aspect of the present invention provides a microgrid power prediction method based on aggregation state constraints, comprising the following steps: Step 1: Obtain multi-source operation data of the microgrid to construct an aggregated state vector; Step 2: Perform multi-scale decomposition on the historical power sequence of the microgrid grid connection point, rearrange the different scale components according to the aggregated state vector, and fuse the rearranged scale components and aggregated state vector to obtain the state-driven input sample. Step 3: Establish a prediction network; input the input samples into the prediction network to obtain the power prediction value for the target time period; Step 4: Based on the aggregated state vector, construct the corresponding set of constraint attributes for the power prediction value, and use the power prediction value and the set of constraint attributes for the target time period as the joint prediction output, and extend it to the target prediction time domain.

[0008] Preferably, in step 1, the aggregated state vector Defined as:

[0009] In the formula, For time period Photovoltaic power output, For time period The overall load power, , Time periods Energy storage and the maximum charging power of electric vehicles, , Time periods Energy storage and the maximum discharge power of electric vehicles For time period The state of charge of the energy storage unit, For time period The number of electric vehicles connected to charging stations. For time period The proportion of fast-charging vehicles, For time period Solar irradiance, For time period Ambient temperature, For time period The intensity of net load fluctuation.

[0010] Preferably, in step 2, the historical power sequence of the microgrid grid connection point is analyzed. Multiscale decomposition can be expressed by the following formula:

[0011] In the formula, For the first Each scale component The number of decomposition layers, This is the residual component.

[0012] Preferably, the number of decomposition layers The center frequency of each scale component is calculated using the following formula. :

[0013] In the formula, Represents time-frequency transformation. For the first The amplitude in the spectrum after time-frequency transformation of each scale component The frequency in the spectrum; The scale components are sorted from high to low according to their center frequencies; When the number of decomposition layers When the center frequency is highest, middle, and lowest, the scale components are defined as high-frequency, mid-frequency, and low-frequency scale components, respectively; when the number of decomposition layers is... When setting the threshold, use the following formula:

[0014] In the formula, For high-frequency segmentation threshold, Thresholds are set for low frequencies. , , These represent the maximum, median, and minimum center frequencies of all scale components, respectively. satisfy The scale components are classified as high-frequency scale components, satisfying... The scale components are classified as mid-frequency scale components, satisfying... The scale components are classified as low-frequency scale components.

[0015] Preferably, in step 2, rearranging the components at different scales based on the aggregated state vector includes: Calculation period No. State-related scores of each scale component It can be expressed by the following formula:

[0016] In the formula, For the first The state sensitivity coefficient vector corresponding to each scale component for transpose, For bias terms, , The number of decomposition layers, It is the aggregated state vector; Based on state-related scores The scale weights are determined using a normalized function, expressed by the following formula:

[0017] In the formula, The first state determined by the aggregated state vector Scale weights of each scale component; According to scale weight For the original scale components Rearrangement is performed using the following formula:

[0018] In the formula, For the first Each re-arrangement scale component.

[0019] Preferably, in step 2, the fusion is performed within the historical window, and the time period is... Input samples Expressed as follows:

[0020] In the formula, For the first Sequence features of each rearranged scale component within a historical window. , The number of decomposition layers, For the length of the history window, This is the aggregated state vector.

[0021] Preferably, in step 4, the set of constraint attributes includes operating mode, fluctuation level, and power feasible boundary, expressed by the following formula:

[0022] In the formula, For time period The set of constraint attributes, For time period The running mode label, For time period The fluctuation level label, For time period The power feasible boundary, where and These represent the lower and upper bounds of the target time period power under the current aggregation state, respectively.

[0023] Preferably, the operating mode is determined based on the current aggregation state and the predicted power level. It can be expressed by the following formula:

[0024] In the formula, This indicates a priority mode for waste disposal. Indicates peak shaving mode. Indicates the economic operating model; For time period Photovoltaic power output, For comprehensive load power, The energy storage unit is in its state of charge. The power forecast value for the target time period; These are the preset load threshold, the threshold under charged state, and the threshold under charged state, respectively. This is the preset power prediction threshold.

[0025] Preferably, the fluctuation level is classified according to the predicted power change rate in adjacent time periods. It can be expressed by the following formula:

[0026] In the formula, and These are the preset lower and upper thresholds for power fluctuation, where 1, 2, and 3 represent low, medium, and high fluctuation levels, respectively. and These are adjacent time periods within the target prediction time domain. t , t +1 power prediction value.

[0027] Preferably, the determination is based on the current load level, photovoltaic output, and flexible resource adjustment capability. It can be expressed by the following formula:

[0028]

[0029] In the formula, For comprehensive load power, For time period Photovoltaic power output, This is the maximum charging power for energy storage. Maximum charging power for electric vehicles To store the maximum discharge power, This represents the maximum discharge power of an electric vehicle.

[0030] A second aspect of the present invention provides a microgrid power prediction system based on aggregation state constraints, for implementing the aforementioned microgrid power prediction method based on aggregation state constraints, comprising: The aggregated state module is used to acquire multi-source operation data of the microgrid to construct an aggregated state vector; The sample construction module is used to perform multi-scale decomposition of the historical power sequence of the microgrid grid connection point, rearrange the different scale components according to the aggregated state vector, and fuse the rearranged scale components and aggregated state vector to obtain the state-driven input sample. The power prediction module is used to build a prediction network and input the input samples into the prediction network to obtain the power prediction value for the target time period. The joint output module is used to construct a set of corresponding constraint attributes for the power prediction value based on the aggregated state vector, and use the power prediction value and the set of constraint attributes for the target time period as the joint prediction output, and extend it to the target prediction time domain.

[0031] Compared with the prior art, the beneficial effects of the present invention include at least the following: The present invention proposes a microgrid power prediction method and system based on aggregated state constraints, which can realize unified modeling and short-term prediction of typical distributed resources such as photovoltaics, loads, electric vehicles and energy storage under multi-source heterogeneous data conditions.

[0032] This invention constructs an aggregated state vector to uniformly represent the access level, load characteristics, energy storage status, and external disturbances of distributed resources at different operating stages, thereby improving the representation of the aggregated operating state of the microgrid by the input sample.

[0033] This invention rearranges power components at different scales based on aggregated state vectors to form state-driven input samples, which can reduce the impact of sample distribution drift under different operating states on prediction results.

[0034] This invention outputs the operating mode, fluctuation level and / or feasible boundary attributes simultaneously with the output power prediction value, so that the prediction results can directly serve the microgrid scheduling, control and operation decision-making.

[0035] This invention can effectively improve the accuracy and interpretability of distributed energy time-series prediction. It can be applied to microgrid hardware-in-the-loop simulation platforms and park-level integrated energy systems, providing technical support for distributed energy scheduling and operation optimization, and has good engineering application prospects. Attached Figure Description

[0036] Figure 1 This is the main flowchart provided according to the embodiments of the present invention; Figure 2 This is the original power curve of the grid connection point in the park provided according to the application example of the present invention; Figure 3 This is a multi-scale decomposition result diagram of grid connection point power provided in accordance with an application example of the present invention; Figure 4 This is a scale weight change curve provided in accordance with the application examples of this invention; Figure 5 This is a power prediction value and feasible boundary joint output diagram provided according to the application example of the present invention; Figure 6 This is a combined output diagram of the operating mode and fluctuation level provided in accordance with the application examples of this invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0038] This invention targets microgrid aggregates composed of distributed resources such as photovoltaics, loads, electric vehicles, and energy storage. First, it acquires multi-source operational data and constructs a state vector representing the aggregated operational state. Then, it performs multi-scale decomposition on historical power sequences and rearranges the components at different scales based on the aggregated state vector to form state-driven input samples. Next, it inputs these input samples into a prediction network, outputting the power prediction result in the target prediction time domain. Finally, it constructs constraint attributes for the prediction output based on the aggregated state, ensuring that the prediction result simultaneously includes the target time period power value and its corresponding operating mode, fluctuation level, and feasible boundary information.

[0039] like Figure 1 As shown, Embodiment 1 of the present invention provides a microgrid power prediction method based on aggregation state constraints, comprising the following steps: Step 1: Obtain multi-source operation data of the microgrid to construct an aggregated state vector.

[0040] In a preferred but non-limiting embodiment of the present invention, step 1 specifically includes: Step 1.1: Obtain operational data of photovoltaic, load, electric vehicle, and energy storage devices in the microgrid, as well as meteorological data such as solar irradiance and ambient temperature. Perform field unification, time alignment, and validity verification on data from different sources to ensure the consistency of subsequent sample construction.

[0041] Step 1.2 involves correcting data with missing, abnormal, or invalid records. In a further preferred but non-limiting embodiment of the invention, short-term missing data is imputed, and abnormal data is filtered out or replaced, thereby obtaining a continuous and valid sequence of operational data.

[0042] Specifically, let the original data sequence be... for:

[0043] In the formula, Represents the observed value at time τ. This represents the number of observation points.

[0044] After correction, a valid data sequence was obtained. :

[0045] In the formula, This represents the corrected data at time τ. To correct the amount of data.

[0046] Step 1.3: Construct an aggregated state vector representing the aggregated operation state of the microgrid based on the multi-source operational data. This aggregated state vector reflects the overall operational state of the distributed resource aggregate in the current time period and serves as the basis for constructing subsequent input samples and generating predicted output constraint attributes. Since the scale of distributed resource access, energy storage status, electric vehicle access behavior, and external disturbances in the microgrid vary significantly across different time periods, the corresponding operating mechanisms and subsequent evolution trends under the same power level are not consistent. Therefore, it is necessary to introduce an aggregated state vector to uniformly represent the current aggregated operational state.

[0047] In one specific embodiment, time period That is, the aggregated state vector of the current prediction period. Defined as:

[0048] In the formula, For time period Photovoltaic power output, For time period The overall load power, , These are the maximum charging power for energy storage and electric vehicles, respectively. , Time periods Energy storage and the maximum discharge power of electric vehicles For time period The state of charge of the energy storage unit, For time period The number of electric vehicles connected to charging stations. For time period The proportion of fast-charging vehicles, For time period Solar irradiance, For time period Ambient temperature, For time period The net load fluctuation intensity. The aggregation state vector It is used both to determine the organization of different scale components in the input sample and to constrain the generation of target output attributes.

[0049] More preferably, the net load fluctuation intensity Expressed as follows:

[0050] In the formula, The previous adjacent time period The overall load power, The previous adjacent time period Photovoltaic power output.

[0051] More preferably, to eliminate the differences in dimensions and value ranges among the components of the aggregated state vector and to enhance the consistency of subsequent state mapping calculations, the state components in the aggregated state vector are normalized to obtain a standardized aggregated state vector. Let the original aggregate state vector be:

[0052] In the formula, , , These are the 1st, 2nd, and 3rd positions in the aggregated state vector, respectively. q Each state component q This represents the number of state components, i.e., the dimension of the aggregated state vector.

[0053] Then the first The normalization result of each state component can be expressed as:

[0054] In the formula, and They represent the first The minimum and maximum values ​​of each state component in the historical samples.

[0055] Therefore, we get:

[0056] In the formula, , , These are the 1st, 2nd, and 3rd positions in the standardized aggregate state vector, respectively. q A standardized state component.

[0057] The standardized aggregated state vector can replace the original aggregated state vector and be used for subsequent rearrangement weight calculation and constraint attribute generation.

[0058] Step 2: Analyze the historical power sequence of the microgrid's grid connection point. Multi-scale decomposition is performed, and the components at different scales are rearranged based on the aggregated state vector. The rearranged scale components and the aggregated state vector are then fused to obtain the state-driven input sample. .

[0059] In a preferred but non-limiting embodiment of the present invention, step 2 specifically includes: Step 2.1 involves performing multi-scale decomposition on the historical power sequence to extract variation features at different time scales. The multi-scale decomposition method can employ empirical mode decomposition, variational mode decomposition, wavelet decomposition, or other methods capable of obtaining multiple scale components.

[0060] More preferably, for the historical power sequence as The decomposition is expressed by the following formula:

[0061] In the formula, For the first Each scale component The number of decomposition layers, This is the residual component.

[0062] More preferably, to distinguish the frequency attributes of components at different scales, the center frequency of each scale component is calculated:

[0063] In the formula, Represents time-frequency transformation. For the first The amplitude in the spectrum after time-frequency transformation of each scale component The frequency is in the spectrum.

[0064] The scale components are sorted from highest to lowest center frequency. Scale components with higher center frequencies represent short-term fluctuations, scale components with intermediate center frequencies represent periodic changes, and scale components with lower center frequencies represent trend changes.

[0065] More preferably, when the number of decomposition layers When the total number of scale components is 3, the scale components with the highest, middle, and lowest center frequencies can be defined as high-frequency, mid-frequency, and low-frequency scale components, respectively; when the number of decomposition layers is... At that time, an adaptive division threshold can be set according to the mean of the center frequencies of different quantiles:

[0066] In the formula, For high-frequency segmentation threshold, Thresholds are set for low frequencies. , , These represent the maximum, median, and minimum center frequencies of all scale components, respectively. They satisfy... The scale components are classified as high-frequency scale components, satisfying... The scale components are classified as mid-frequency scale components, satisfying... The scale components are classified as low-frequency scale components.

[0067] It is worth noting that the decomposition according to three scales—high frequency, medium frequency, and low frequency—is only used to illustrate one implementation of the present invention and does not constitute a limitation on the multi-scale decomposition method or the number of decomposition layers.

[0068] Step 2.2, based on the aggregated state vector The components at each scale are correlated and arranged to form different input organization methods under different operating conditions.

[0069] It is understandable that, since the contribution of each scale component of the historical power sequence to the future power evolution varies under different aggregation states, it is difficult to reflect the impact of state changes on the effectiveness of information at different scales if a fixed input organization method is used. Therefore, it is necessary to rearrange the scale components based on the aggregation state vector.

[0070] Let the first state after aggregation be constrained. Each rearrangement scale component is The rearrangement relationship is expressed by the following formula:

[0071] In the formula, , The number of decomposition layers, This represents a mapping function that sorts, filters, weights, or groups scale components based on the aggregated state vector.

[0072] In a further preferred but non-limiting embodiment of the present invention, step 2.2 specifically includes: Step 2.2.1, based on the aggregated state vector Calculate the state correlation score for each scale component. (The time period is not specified.) No. State-related scores of each scale component Expressed as follows:

[0073] In the formula, For the first The state sensitivity coefficient vector corresponding to each scale component for transpose, This is a bias term. More preferably, and It is obtained through offline training based on historical samples; or it is preset by human based on the statistical results of historical operation data and used as a fixed parameter during the operation of the prediction model.

[0074] Step 2.2.2, based on the state-related score The scale weights are determined using a normalization function. It can be expressed by the following formula:

[0075] in, satisfy:

[0076] Step 2.2.3, based on scale weights For the original scale components The rearrangement scale components are obtained by weighted calculation after re-arranging. It can be expressed by the following formula:

[0077] In the formula, The first state determined by the aggregated state vector Scale weights of each scale component.

[0078] In one specific embodiment, high-frequency scale components It is quite sensitive to the state of charge of energy storage and the number of electric vehicles connected. middle and The corresponding weight is relatively high, and The weight is negative. When the energy storage state of charge is low and the number of electric vehicles connected is large, the microgrid's flexible regulation capability is weak and the risk of short-term impact is high. At this time, the high-frequency scale component... The future power evolution plays a dominant role, and correspondingly, the aggregated state vector via state sensitivity coefficient vector After mapping, high-frequency scale components can be made Corresponding state-related score Increase, thereby increasing the weight of high-frequency scale components. .

[0079] In another specific embodiment, the low-frequency scale component It is quite sensitive to photovoltaic power output and weather disturbances. middle , and The corresponding weights are relatively high. When photovoltaic output is high and meteorological disturbances are small, the system operation is relatively stable, and the low-frequency scale components are relatively low. The future power evolution plays a dominant role, and correspondingly, the aggregated state vector via state sensitivity coefficient vector After mapping, the low-frequency scale components can be made... Corresponding state-related score Increase, thereby increasing the weight of low-frequency scale components. .

[0080] Step 2.3: Within the historical window, fuse the multi-scale components and the aggregated state vector to generate state-driven input samples. Let the length of the historical window be... Then the time period Corresponding input samples Expressed as follows:

[0081] In the formula, For the first Sequence features of each rearranged scale component within a historical window. This is the aggregated state vector.

[0082] Understandably, under different aggregation states, the combination, feature ordering, or constraint relationships of the components at each scale in the input sample differ, so that the input sample can reflect the impact of changes in the operating state of distributed resource aggregation on the historical power sequence. The rearranged state-driven input sample can characterize the differentiated effect of historical power information at each scale on future power evolution under different aggregation operating states, reduce the impact of input sample distribution drift, and further improve prediction accuracy.

[0083] Step 3, build the prediction network; use state-driven input samples Input the data into the prediction network to obtain the power prediction value for the target time period. .

[0084] In a preferred but non-limiting embodiment of the present invention, the prediction network is implemented using a non-autoregressive structure to support the parallel output of the power sequence of the target time period, or it is implemented using other deep learning networks suitable for time-series mapping.

[0085] More preferably, the prediction network includes a feature encoding module and an output mapping module. The feature encoding module is used to extract temporal features from the state-driven input samples, and the output mapping module is used to generate power prediction values ​​for the target time period.

[0086] Step 4, based on the aggregated state vector , is the predicted power value Construct the corresponding set of constraint attributes The power forecast value for the target period and constraint attribute set Together as joint prediction output And extend it to the target prediction time domain.

[0087] In a preferred but non-limiting embodiment of the present invention, the set of constraint attributes includes: operating mode, fluctuation level, and power feasible boundary, expressed by the following formula:

[0088] In the formula, For time period The set of constraint attributes, For time period The operating mode label is used to characterize the typical operating category of the microgrid aggregate during a target time period. For time period The fluctuation level label is used to characterize the severity of power changes. For time period The power feasible boundary is used to characterize the feasible range of power for the target time period under the current aggregation state.

[0089] The set of constraint attributes It can be determined by the aggregated state vector and the predicted power value, as expressed by the following formula:

[0090] In the formula, Indicates based on the aggregated state vector and predicted power value A mapping function that generates a set of constraint attributes.

[0091] In a preferred but non-limiting embodiment of the present invention, step 4 specifically includes: Step 4.1: Determine the operating mode based on the current aggregation status and predicted power level. More preferably, the operating mode determination rule is expressed by the following formula:

[0092] In the formula, This indicates a priority mode for waste disposal. Indicates peak shaving mode. Indicates the economic operating model; These are the preset load threshold, the threshold under charged state, and the threshold under charged state, respectively. This is the preset power prediction threshold.

[0093] Among them, the priority mode for grid connection corresponds to the operation scenario where the photovoltaic output is high and the energy storage is in a rechargeable state; the peak shaving mode corresponds to the operation scenario where the load is high and the energy storage has the ability to discharge; and the economic operation corresponds to the normal operation scenario where the grid connection point power does not trigger the peak shaving constraint and the photovoltaic grid connection conditions are not dominant, with the goal of optimizing the overall energy cost.

[0094] Step 4.2: Classify fluctuation levels based on the predicted power change rate in adjacent time periods. It can be expressed by the following formula:

[0095] In the formula, and The preset lower threshold for power fluctuation, upper threshold for power fluctuation, and fluctuation level. The values ​​1, 2, and 3 correspond to the low, medium, and high fluctuation levels, respectively. and These are adjacent time periods within the target prediction time domain. t , t +1 power prediction value.

[0096] Step 4.3: Obtain the power feasible boundary by mapping the aggregated state vector. It can be expressed by the following formula:

[0097] Determine the feasible power boundary based on the current load level, photovoltaic output, and flexible resource adjustment capabilities. It can be expressed by the following formula:

[0098]

[0099] In the formula, This represents the mapping function that determines the power feasible boundary from the aggregated state vector. and These represent the lower and upper bounds of the target time period power under the current aggregation state, respectively. This is the maximum charging power for energy storage. Maximum charging power for electric vehicles To store the maximum discharge power, This represents the maximum discharge power of an electric vehicle.

[0100] Step 4.4, combine the power prediction value for the target time period. To form a joint prediction output It can be expressed by the following formula:

[0101] In the formula, The power forecast value for the target time period. This is the set of constraint attributes corresponding to the predicted power value.

[0102] It is understood that by adding constraint attributes to describe the operating state information corresponding to the predicted power results, the present invention makes the predicted power results not limited to a single power point value, thereby improving the usability of the predicted power results.

[0103] Step 4.5 extends the single-time output format from Step 4.4 to the target prediction time domain, expressed by the following formula:

[0104] In the formula, each constraint set All of these include operating modes, fluctuation levels, and power feasibility boundaries.

[0105] For example, if the target prediction time domain is the next 24 hours, the joint prediction output includes hourly power prediction curves and corresponding operating modes, fluctuation levels and feasible boundary information.

[0106] It is understandable that the improvement of this invention compared to traditional prediction methods lies in the changes in the way the input samples are constructed and the way the output is organized.

[0107] Traditional methods typically involve directly splicing together historical power sequences, meteorological data, and load data, with all samples input into the prediction network using the same structure. This approach fails to differentiate between different aggregation states, such as "high photovoltaic output, low SOC of energy storage, concentrated electric vehicle access, and drastic load fluctuations," easily leading to samples from different operating mechanisms being mixed in the same input representation, resulting in sample distribution drift and unstable prediction results.

[0108] This invention first constructs an aggregated state vector, then uses this state vector to calculate the state-related weights of components at different scales, and rearranges the multi-scale power components. In this way, the input samples are not fixedly spliced ​​together, and the importance of information at different scales can be adjusted as the aggregated state of the microgrid changes. For example, when local fluctuations are large or the flexible regulation capability is weak, the weight of high-frequency components is increased; when the operating state is relatively stable, the weight of low-frequency components is increased. This processing can enhance the consistency between the input samples and the current aggregated operating state, reduce the interference of irrelevant scale components on the prediction results, and improve the applicability and stability of the model under different operating states.

[0109] Furthermore, traditional methods typically only output power point prediction values. This invention, while outputting power prediction values, simultaneously generates operating modes, fluctuation levels, and feasible power boundaries. This ensures that the prediction results not only provide future power forecasts but also indicate the operating state under which the prediction occurs, the level of fluctuation risk, and the acceptable power boundaries. This allows the prediction results to directly serve microgrid dispatching, peak shaving control, renewable energy integration assessment, and safety boundary verification, improving the engineering usability of the prediction results.

[0110] Embodiment 2 of the present invention provides a microgrid power prediction system based on aggregation state constraints, used to implement the microgrid power prediction method based on aggregation state constraints described in Embodiment 1, comprising: The aggregated state module is used to acquire multi-source operation data of the microgrid to construct an aggregated state vector; The sample construction module is used to perform multi-scale decomposition of the historical power sequence of the microgrid grid connection point, rearrange the different scale components according to the aggregated state vector, and fuse the rearranged scale components and aggregated state vector to obtain the state-driven input sample. The power prediction module is used to build a prediction network and input the input samples into the prediction network to obtain the power prediction value for the target time period. The joint output module is used to construct a set of corresponding constraint attributes for the power prediction value based on the aggregated state vector, and use the power prediction value and the set of constraint attributes for the target time period as the joint prediction output, and extend it to the target prediction time domain.

[0111] To further illustrate the technical solution of this invention and its beneficial effects, an application example is provided below. In one application example, the microgrid power prediction method based on aggregation state constraints described in Example 1 is applied to a park-level photovoltaic-storage-charging microgrid scenario to predict the power of a certain park microgrid.

[0112] The park's microgrid includes distributed photovoltaic (PV) power generation units, energy storage units, electric vehicle charging facilities, comprehensive park loads, and meteorological information collection units. The distributed PV power generation units provide renewable energy output, the energy storage units smooth power fluctuations and participate in energy regulation, the electric vehicle charging facilities reflect the access and charging behavior of electric vehicles within the park, and the comprehensive park loads include office loads, air conditioning loads, and general power loads.

[0113] In this scenario, the present invention predicts the power evolution in the future target prediction time domain based on the current operating state of the microgrid aggregate, and simultaneously provides the corresponding operating mode, fluctuation level, and power feasible boundary to support the operation and scheduling of park-level microgrids. The specific prediction process and prediction results are as follows.

[0114] like Figure 2 As shown, the raw power curve of a microgrid in a certain park reflects the net power change of the microgrid aggregate during a continuous sampling period. In the figure, the vertical axis P pcc The curve represents the net power at the microgrid's grid connection point, with the horizontal axis representing time. This curve simultaneously incorporates load changes, photovoltaic power output fluctuations, electric vehicle access behavior, and the impact of energy storage regulation, exhibiting significant non-stationarity and short-term fluctuation characteristics. Therefore, further multi-scale decomposition and state-constraint modeling are required.

[0115] The historical power sequence of the microgrid's grid connection point in the industrial park was decomposed using the empirical mode decomposition method, yielding several intrinsic mode function (IMF) components. The first three IMF components were selected as multi-scale components.

[0116] like Figure 3 As shown, , , Arranged from shortest to longest local time scale, they are used to characterize high-frequency short-term fluctuations, mid-frequency periodic changes, and low-frequency trend changes, respectively.

[0117] like Figure 4 As shown, the scale weights corresponding to different scale components , , The scale weights change with the aggregated state vector. When the microgrid's operating state changes, the scale weights are adjusted synchronously, thereby altering the contribution of high-frequency, mid-frequency, and low-frequency components to the input samples.

[0118] like Figure 5 As shown, the prediction network outputs the power prediction value in the target prediction time domain and simultaneously provides the power feasible boundary. In the figure, Actual represents the actual power curve, Forecast represents the power prediction curve, and P... max As the upper bound for power prediction, P min As the lower bound for power prediction, P max and P min The curves together form the power feasible boundary. The power feasible boundary is determined by the current load level, photovoltaic output, and the adjustability of flexible resources such as energy storage and electric vehicles. It is used to characterize the allowable range of variation in the net power of the grid-connected point during the target period under the current aggregation state. Figure 5 It displays the joint output of the power prediction value and the corresponding power feasible boundary in the target prediction time domain, which can be used for boundary verification in scheduling control.

[0119] like Figure 6 As shown, based on the predicted power output, the operating mode and fluctuation level are further output. In the figure, the vertical axis "Mode" in the upper output represents the operating mode, with values ​​of 1, 2, and 3 corresponding to priority absorption mode, peak shaving mode, and economic operation mode, respectively. In the lower output, the vertical axis "Level" represents the fluctuation level, with values ​​of 1, 2, and 3 corresponding to low fluctuation level, medium fluctuation level, and high fluctuation level, respectively. The operating mode is used to distinguish typical states such as priority absorption, peak shaving, or economic operation, while the fluctuation level reflects the severity of power changes within the target prediction time domain. Through this joint output method, the prediction result includes not only power values ​​but also operational attribute information for scheduling decisions.

[0120] This demonstrates that the present invention can not only output the power prediction value in the target prediction time domain, but also simultaneously output the operation attribute information corresponding to the prediction value.

[0121] Compared with the prior art, the beneficial effects of the present invention include at least the following: The present invention proposes a microgrid power prediction method and system based on aggregated state constraints, which can realize unified modeling and short-term prediction of typical distributed resources such as photovoltaics, loads, electric vehicles and energy storage under multi-source heterogeneous data conditions.

[0122] This invention constructs an aggregated state vector to uniformly represent the access level, load characteristics, energy storage status, and external disturbances of distributed resources at different operating stages, thereby improving the ability of input samples to reflect the aggregated operating state of microgrids.

[0123] This invention rearranges power components at different scales based on aggregated state vectors to form state-driven input samples, which can reduce the impact of sample distribution drift under different operating states on prediction results.

[0124] This invention outputs the operating mode, fluctuation level and / or feasible boundary attributes simultaneously with the output power prediction value, so that the prediction results can directly serve the microgrid scheduling, control and operation decision-making.

[0125] This invention can effectively improve the accuracy and interpretability of distributed energy time-series prediction. It can be applied to microgrid hardware-in-the-loop simulation platforms and park-level integrated energy systems, providing technical support for distributed energy scheduling and operation optimization, and has good engineering application prospects.

[0126] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0127] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A microgrid power prediction method based on aggregation state constraints, characterized in that, Includes the following steps: Step 1: Obtain multi-source operation data of the microgrid to construct an aggregated state vector; Step 2: Perform multi-scale decomposition on the historical power sequence of the microgrid grid connection point, rearrange the different scale components according to the aggregated state vector, and fuse the rearranged scale components and aggregated state vector to obtain the state-driven input sample. Step 3: Establish a prediction network; input the input samples into the prediction network to obtain the power prediction value for the target time period; Step 4: Based on the aggregated state vector, construct the corresponding set of constraint attributes for the power prediction value, and use the power prediction value and the set of constraint attributes for the target time period as the joint prediction output, and extend it to the target prediction time domain.

2. The microgrid power prediction method based on aggregation state constraints as described in claim 1, characterized in that: In step 1, aggregate the state vector Defined as: In the formula, For time period Photovoltaic power output, For time period The overall load power, , Time periods Energy storage and the maximum charging power of electric vehicles, , Time periods Energy storage and the maximum discharge power of electric vehicles, For time period The state of charge of the energy storage unit, For time period The number of electric vehicles connected to charging stations. For time period The proportion of fast-charging vehicles, For time period Solar irradiance, For time period Ambient temperature, For time period The intensity of net load fluctuation.

3. The microgrid power prediction method based on aggregation state constraints as described in claim 1, characterized in that: In step 2, the historical power sequence of the microgrid grid connection point is analyzed. Multiscale decomposition can be expressed by the following formula: In the formula, For the first Each scale component The number of decomposition layers, This is the residual component.

4. The microgrid power prediction method based on aggregation state constraints as described in claim 3, characterized in that: Decomposition layer number The center frequency of each scale component is calculated using the following formula. : In the formula, Represents time-frequency transformation. For the first The amplitude in the spectrum after time-frequency transformation of each scale component The frequency in the spectrum; The scale components are sorted from high to low according to their center frequencies; When the number of decomposition layers When the center frequency is highest, middle, and lowest, the scale components are defined as high-frequency, mid-frequency, and low-frequency scale components, respectively. When the number of decomposition layers When setting the threshold, use the following formula: In the formula, For high-frequency segmentation threshold, Thresholds are set for low frequencies. , , These represent the maximum, median, and minimum center frequencies of all scale components, respectively. satisfy The scale components are classified as high-frequency scale components, satisfying... The scale components are classified as mid-frequency scale components, satisfying... The scale components are classified as low-frequency scale components.

5. The microgrid power prediction method based on aggregation state constraints as described in claim 1, characterized in that: Step 2, which involves rearranging the components at different scales based on the aggregated state vector, includes: Calculation period No. State-related scores of each scale component It can be expressed by the following formula: In the formula, For the first The state sensitivity coefficient vector corresponding to each scale component for transpose, For bias terms, , The number of decomposition layers, It is the aggregated state vector; Based on state-related scores The scale weights are determined using a normalized function, expressed by the following formula: In the formula, The first state determined by the aggregated state vector Scale weights of each scale component; According to scale weight For the original scale components Rearrangement is performed using the following formula: In the formula, For the first Each re-arrangement scale component.

6. The microgrid power prediction method based on aggregation state constraints as described in claim 1, characterized in that: In step 2, the data is merged within the historical window, by time period. Input samples Expressed as follows: In the formula, For the first Sequence features of each rearranged scale component within a historical window. , The number of decomposition layers, For the length of the history window, This is the aggregated state vector.

7. The microgrid power prediction method based on aggregation state constraints as described in claim 1, characterized in that: In step 4, the set of constraint attributes includes the operating mode, fluctuation level, and power feasible boundary, expressed by the following formula: In the formula, For time period The set of constraint attributes, For time period The running mode label, For time period The fluctuation level label, For time period The power feasible boundary, where and These represent the lower and upper bounds of the target time period power under the current aggregation state, respectively.

8. The microgrid power prediction method based on aggregation state constraints as described in claim 7, characterized in that: Determine the operating mode based on the current aggregation status and predicted power level. It can be expressed by the following formula: In the formula, This indicates a priority mode for waste disposal. Indicates peak shaving mode. Indicates the economic operating model; For time period Photovoltaic power output, For comprehensive load power, The energy storage unit is in its state of charge. The power forecast value for the target time period; These are the preset load threshold, the threshold under charged state, and the threshold under charged state, respectively. This is the preset power prediction threshold.

9. The microgrid power prediction method based on aggregation state constraints as described in claim 7, characterized in that: Fluctuation levels are classified based on the predicted power change rate in adjacent time periods. It can be expressed by the following formula: In the formula, and These are the preset lower and upper thresholds for power fluctuation, where 1, 2, and 3 represent low, medium, and high fluctuation levels, respectively. and These are adjacent time periods within the target prediction time domain. t , t +1 power prediction value.

10. The microgrid power prediction method based on aggregation state constraints as described in claim 7, characterized in that: Determined based on current load levels, photovoltaic output, and flexible resource adjustment capabilities. It can be expressed by the following formula: In the formula, For comprehensive load power, For time period Photovoltaic power output, This is the maximum charging power for energy storage. Maximum charging power for electric vehicles To store the maximum discharge power, This represents the maximum discharge power of an electric vehicle.

11. A microgrid power prediction system based on aggregation state constraints, used to implement the microgrid power prediction method based on aggregation state constraints according to any one of claims 1-10, characterized in that, include: The aggregated state module is used to acquire multi-source operation data of the microgrid to construct an aggregated state vector; The sample construction module is used to perform multi-scale decomposition of the historical power sequence of the microgrid grid connection point, rearrange the different scale components according to the aggregated state vector, and fuse the rearranged scale components and aggregated state vector to obtain the state-driven input sample. The power prediction module is used to build a prediction network and input the input samples into the prediction network to obtain the power prediction value for the target time period. The joint output module is used to construct a set of corresponding constraint attributes for the power prediction value based on the aggregated state vector, and use the power prediction value and the set of constraint attributes for the target time period as the joint prediction output, and extend it to the target prediction time domain.

Images