Outdoor portable energy storage power supply intelligent management system
The outdoor portable energy storage power supply intelligent management system, which combines multi-dimensional data acquisition and LSTM model with swarm intelligence optimization algorithm, solves the problems of single data acquisition and inaccurate energy consumption prediction, and achieves highly reliable energy regulation in emergency rescue scenarios to ensure continuous power supply to core loads.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CSSC HAISHEN MEDICAL TECH CO LTD
- Filing Date
- 2025-12-26
- Publication Date
- 2026-05-15
AI Technical Summary
Existing intelligent management systems for portable outdoor energy storage power supplies suffer from problems such as limited data collection dimensions, inaccurate energy consumption prediction, and lack of targeted energy regulation in emergency rescue scenarios. They cannot meet high reliability requirements, especially in complex and ever-changing emergency situations, and cannot guarantee continuous power supply to core loads.
By collecting multi-dimensional data (battery status, interface status, environmental status) and combining it with timestamp synchronization, an LSTM model is used to predict energy consumption. Hyperparameters are optimized by combining swarm intelligence optimization algorithms to calculate the solar energy production range, and scene marking and load classification strategies are implemented to achieve precise energy regulation.
It achieves complete acquisition of multi-dimensional data and high-precision energy consumption prediction, dynamically matches energy supply and demand, ensures stable power supply to core loads, and improves the energy utilization and reliability of the system.
Smart Images
Figure CN122052265A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent management system for a portable outdoor energy storage power supply. Background Technology
[0002] Portable outdoor energy storage power supplies are critical energy backup devices for on-site emergency rescue, responsible for providing a continuous and stable power supply to various emergency medical equipment, communication equipment, and lighting tools. In emergency rescue scenarios, the level of intelligent management of the power system directly determines the operational reliability of the core load, and thus the safety of the injured. However, existing intelligent management solutions for portable energy storage power supplies used in such scenarios have significant shortcomings and are unable to meet the high reliability requirements of emergency rescue:
[0003] Firstly, existing solutions rely on a single data acquisition dimension, mostly monitoring only one of the battery's own status or environmental parameters, failing to effectively integrate the three-dimensional characteristic data of battery status, load (interface) conditions, and environmental conditions. This results in incomplete system perception capabilities, failing to provide comprehensive data support for accurate energy consumption prediction and energy scheduling, and creating hidden dangers in complex and ever-changing emergency situations.
[0004] Secondly, existing energy consumption prediction models are overly simplified, failing to fully exploit the inherent temporal correlations and dynamic changes in electrical loads during emergency operations. Furthermore, the model hyperparameter configuration relies on human experience, resulting in low accuracy in predicting short-term energy consumption. This leads to energy regulation decisions based on erroneous expectations, making it impossible to achieve precise "predictive" power allocation.
[0005] Third, and most importantly, the existing solution fails to effectively integrate the fluctuating capacity forecasting of on-site renewable energy sources such as solar power, relying solely on passive regulation based on remaining battery power. This results in low overall energy utilization during prolonged rescue missions, and in critical situations where battery power is scarce and solar resupply is uncertain, it cannot intelligently and reliably prioritize the continuous power supply to core loads directly related to patient life safety, such as monitors and ventilators, posing a risk of core equipment outages due to energy supply imbalances.
[0006] Therefore, for emergency rescue, a professional field with extremely stringent requirements for power supply, there is an urgent need for a portable intelligent energy storage power management system that can achieve multi-dimensional data fusion, high-precision energy consumption prediction, and intelligent coordination between batteries and renewable energy sources, thereby ensuring the absolute stability of power supply to core loads. Summary of the Invention
[0007] To address the shortcomings of existing technologies, this invention provides an intelligent management system for portable outdoor energy storage power supplies, which solves the problems of insufficient data support, inaccurate energy consumption prediction, and lack of targeted regulation in existing outdoor energy storage power supplies.
[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent management system for portable outdoor energy storage power supplies, comprising:
[0009] The data acquisition module collects the remaining SOC and charge / discharge rate of the battery pack from the battery status dimension, the total energy consumption of the interface and the number of interfaces working in real time from the interface status dimension, and the ambient temperature, light intensity and airflow speed from the environmental status dimension at fixed time intervals of T minutes. The three types of data are aligned using timestamp synchronization technology to obtain a three-dimensional feature dataset.
[0010] The data processing module, based on a three-dimensional feature dataset, uses the remaining power SOC, charge and discharge rate, number of interfaces working in real time, ambient temperature, light intensity, and airflow speed as inputs to the LSTM and the total energy consumption of the interfaces as the output of the LSTM. After training, the optimal LSTM total energy consumption prediction model for the interfaces is obtained.
[0011] The energy regulation module collects the current remaining power SOC, charge / discharge rate, number of interfaces in real time, ambient temperature, light intensity, and airflow speed, and substitutes them into the optimal LSTM interface total energy consumption prediction model to predict the total power consumption of the interface at the next moment. At the same time, it calculates the solar energy production range based on the currently collected light intensity and the time interval T between the current moment and the next moment, and compares the predicted total power consumption of the interface at the next moment with the solar energy production range to mark the scene, thereby adopting different energy regulation strategies.
[0012] As a further aspect of the present invention, the specific method for collecting the number of interfaces in real time is as follows: assign a unique physical identifier to each interface, and detect whether there is current output in real time through the miniature current sensor built into the interface; if the current value of a certain interface is greater than 0 for three consecutive times, the interface is determined to be in working state; count the number of interfaces in all working states and take them as the number of interfaces in real time; wherein, the time for the interface to collect the current value three times consecutively needs to be within a fixed time interval T.
[0013] As a further aspect of the present invention, the specific operation for training the LSTM interface total energy consumption prediction model is as follows:
[0014] All input and output features are normalized by min-max and mapped to the [0,1] interval. At the same time, continuous feature data are reconstructed into time series samples according to timestamps. The time step is set to Ts to form input time series sequence-output label sample pairs. The time series samples are divided into training set, validation set and test set in a ratio of 7:2:1.
[0015] Build a four-layer architecture for the LSTM model, consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer, and set the corresponding parameters.
[0016] The training set is substituted into the already built LSTM model for training, and the mean squared error (MSE) is selected as the loss function. At the same time, the Adam optimizer is configured to perform backpropagation updates of the model parameters.
[0017] After each training iteration, the MSE loss of the current model parameters is immediately evaluated using the validation set. The validation set MSE is used as the monitoring indicator and the patience value is set to 3. That is, when the validation set loss does not decrease for 3 consecutive rounds, the model training is automatically terminated. At the same time, the model weights with the lowest validation set loss during the training process are saved and marked as the optimal model weights.
[0018] Load the optimal model weights saved during training to obtain the optimal LSTM interface total energy consumption prediction model, and perform inference calculations on the test set to calculate the root mean square error (RMSE) and mean absolute percentage error (MAPE).
[0019] As a further embodiment of the present invention, the specific parameters corresponding to the four layers are as follows: the dimension of the input layer is (Ts, 6); the number of hidden units in the first LSTM layer is 64, the return sequence function is enabled, and the activation function is tanh; the number of hidden units in the second LSTM layer is 32, the return sequence function is not enabled, and the activation function is tanh; the number of neurons in the fully connected layer is 16, and the activation function is ReLU; the number of neurons in the output layer is 1; the batch size is 32, the learning rate is 0.001, the number of iterations is 100, and the dropout rate is 0.2.
[0020] As a further aspect of the present invention, a swarm intelligence optimization algorithm is used to optimize the time step Ts, the number of hidden units in two LSTM layers, and the learning rate of the LSTM model. The swarm intelligence optimization algorithm is one of the following: mirage search optimization algorithm, RRT-based optimizer, and cloud drift optimization algorithm.
[0021] As a further aspect of the present invention, the specific steps for calculating the solar energy production range are as follows:
[0022] The single-step prediction time interval of the optimal LSTM interface total energy consumption prediction model is denoted as T, which is consistent with the fixed time interval. The current light intensity is denoted as L0, and the current airflow speed is denoted as v.
[0023] Based on the solar panel's conversion efficiency α and panel area S, according to formula P base =L0×ɳ×S÷10 6 Calculate the basic output power P base Set a fluctuation coefficient k and calculate the upper limit P of the basic production capacity. up =P base ×(1+k), lower limit P low =P base×(1-k), we obtain the basic range of solar energy production capacity as [P low ×T÷60,P up [×T÷60];
[0024] The corrected solar panel conversion efficiency ɳ' is calculated using the formula ɳ'=ɳ+(v-v0)×△ɳ, where v o The standard airflow velocity is the most suitable for solar panels, and Δɳ is the change in the conversion efficiency of the solar panels.
[0025] Substitute ɳ' into P base =L0×ɳ×S÷10 6 Calculate P up '、P low ', thus obtaining the final solar energy production capacity range E solar =[P low '×T÷60,P up '×T÷60'.
[0026] As a further aspect of the present invention, the specific steps for scene marking are as follows:
[0027] The battery's remaining state of charge (SOC) is converted into actual remaining energy (E) using an extended Kalman filter algorithm. bat The specific formula is E bat =Q×SOC, and then compared with the solar energy capacity range E. solar Combining these, we obtain the fundamental energy range E. base =[E solar_low +E bat E solar_high +E bat Where Q is the rated capacity of the battery, and E is the rated capacity of the battery. solar_low E solar_high Solar energy capacity range E solar The lower and upper limits;
[0028] Collect the current battery charge / discharge rate C rate If the device is in a charging state, calculate the additional charging energy E within time T. extra =C rate ×Q×T÷60, E extra Superimposed on E available The upper and lower limits are used to obtain the final usable energy range E. available =[E base_low +E extra E extra_high +E extra ], where E base_low E extra_high Basic energy range E base The lower and upper limits;
[0029] Invoke the trained optimal LSTM interface total energy consumption prediction model and output the predicted interface total energy consumption value E for the next time point. pred Based on the model's error distribution on the test set, it outputs a 95% confidence interval E for prediction. pred_conf =[E pred -△E,E pred +△E], where △E is the critical error value;
[0030] Determine E pred_conf With E available The overlapping interval [E] overlap_min E overlap_max ], where E overlap_min =max(E pred -△E,E base_low +E extra E overlap_max =min(E pred +△E,E extra_high +E extra );
[0031] If E overlap_min ≤E overlap_max Then calculate the overlap R=(E overlap_max -E overlap_min ) / (E pred +△E-(E pred -△E)); If E overlap_min >E overlap_max If R = 0, then R = 0;
[0032] Calculate E available The midpoint value E available_mid The predicted total power consumption E of the interface is obtained. pred With midpoint value E available_mid The deviation ratio D = |E pred -E available_mid | / E available_mid ;
[0033] The scene is labeled according to R and D: if R≥R1 and D≤D1, it is labeled as scene one; if R2≤R<R1 or D1<D≤D2, it is labeled as scene two; if R<R2 or D>D2, it is labeled as scene three. Here, R1 and R2 are the lower and upper limits of the overlap R, and D1 and D2 are the lower and upper limits of the deviation ratio D.
[0034] As a further aspect of the present invention, the method for determining the value of △E is as follows: calculate the single-step prediction error of the optimal LSTM interface total energy consumption prediction model on all samples in the test set, obtain the average value μ and standard deviation σ of the data set, and calculate the critical error value △E according to the formula △E=1.96×σ+|μ|.
[0035] As a further aspect of the present invention, adopting different energy regulation strategies based on the labeled scenario specifically includes:
[0036] If it is Scenario 1, the following energy regulation strategy will be implemented: The functional operation status of the load corresponding to all interfaces will be collected in real time through the unified communication protocol to confirm that the core loads are in normal working mode; at the same time, for non-core loads, the corresponding real-time power benchmark value will be derived based on the preset energy consumption benchmark value within T time period, that is, power benchmark value = energy consumption benchmark value ÷ T, and the current operating power of non-core loads will be locked at this benchmark value to prevent them from increasing their power on their own.
[0037] If it is scenario two, the following energy regulation strategy will be implemented: Based on the total energy consumption benchmark of the interface within the current time T, the corresponding real-time total power benchmark will be derived; the total power that can be reduced for all non-core loads will be counted in real time to ensure that its proportion of the current real-time total power of the interface is ≤15%; at the same time, the adjusted status will be fed back in real time through the unified communication protocol to confirm that the operating power of the core loads is stable.
[0038] If it is scenario three, then the following energy regulation strategy will be implemented: based on the current available energy range E available Predicted energy consumption E for the next T cycle pred Derivation of the core load priority energy consumption quota = E within time T. available The lower limit is multiplied by 80% and converted into a real-time power benchmark to ensure that more than 80% of the battery's output energy is supplied to the core load through this power benchmark. At the same time, emergency power replenishment prompts are pushed through the human-machine interface of the portable power supply to guide users to connect backup lithium battery modules or adjust the orientation of solar panels to increase production capacity until the matching degree between the available energy range and the predicted energy consumption returns to the range of Scenario 1 or Scenario 2.
[0039] As a further aspect of the present invention, the core load refers to a load device that is directly related to the patient's life safety, supports the core diagnosis and treatment actions of outdoor emergency care, and cannot be temporarily interrupted or have its power reduced; the non-core load refers to a load device that provides auxiliary convenience for outdoor emergency care, does not directly participate in the core diagnosis and treatment actions, and will not threaten the patient's life safety if its power is interrupted or reduced.
[0040] This invention provides an intelligent management system for portable outdoor energy storage power supplies, which has the following advantages compared with existing technologies:
[0041] (1) This invention, through the collaborative acquisition of three-dimensional features of battery, interface and environment, combined with timestamp synchronization technology, makes up for the defects of single data dimension and misaligned time sequence in the existing solution, and provides a complete data source for energy consumption prediction and energy regulation.
[0042] (2) This invention uses the LSTM model to mine temporal correlation features, combines it with the swarm intelligence optimization algorithm to optimize hyperparameters, and combines the early stopping mechanism and dual index evaluation to solve the problems of blind hyperparameter configuration and large prediction error in traditional models, and achieves accurate prediction of total interface energy consumption.
[0043] (3) This invention achieves dynamic matching of energy supply and energy consumption demand through solar energy capacity range calculation, scenario marking and load classification strategy. While prioritizing the operation of core loads, it makes reasonable use of renewable energy and redundant energy, solving the problems of lack of scenario adaptation and low energy utilization in existing regulation. Attached Figure Description
[0044] Figure 1 This is the system principle block diagram of the present invention;
[0045] Figure 2 A flowchart illustrating the scene marking steps of this invention. Detailed Implementation
[0046] 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.
[0047] Example 1
[0048] like Figure 1 This invention provides an intelligent management system for portable outdoor energy storage power supplies, comprising:
[0049] The data acquisition module collects the remaining SOC and charge / discharge rate of the battery pack from the battery status dimension, the total energy consumption of the interface and the number of interfaces working in real time from the interface status dimension, and the ambient temperature, light intensity and airflow speed from the environmental status dimension at fixed time intervals of T minutes. The three types of data are aligned using timestamp synchronization technology to obtain a three-dimensional feature dataset.
[0050] In outdoor emergency rescue scenarios, some load devices have low power adaptive functions, such as breathing modules and monitoring modules. When the battery SOC is lower than a certain set range, it will automatically reduce the power consumption of non-core functions, resulting in a decrease in the load power consumption of the corresponding interface. Conversely, when the load is at a high SOC, it will maintain full-function operation, and the load power consumption of the corresponding interface will be stable.
[0051] The charge / discharge rate is the ratio of the real-time charge / discharge current to the battery's rated capacity. It directly reflects the energy consumption level of the load at the corresponding interface at the current time. In outdoor emergency rescue scenarios, the energy consumption of the load at the corresponding interface has a short-cycle continuity. For example, when the defibrillator enters the charging stage, the discharge rate will increase sharply, and this high-rate state will continue for a period of time.
[0052] Total interface energy consumption is the sum of the load energy consumption of all interfaces within a fixed time interval T. Under normal operating conditions, total interface energy consumption fluctuates regularly: during the day, the total interface energy consumption is low because the lighting load is turned off, while at night, the total interface energy consumption increases because the lighting load is turned on.
[0053] In outdoor emergency rescue scenarios, each working interface corresponds to an independent load, such as a ventilator, lighting equipment, or charger. Moreover, the loads are mostly in parallel operation mode, meaning that the energy consumption of multiple interfaces corresponding to the loads accumulates within the same time interval, directly increasing the total energy consumption of the interfaces.
[0054] The specific method for collecting the number of interfaces in real-time operation is as follows: assign a unique physical identifier to each interface, and detect whether there is current output in real time through the miniature current sensor built into the interface; if the current value of an interface is greater than 0 for 3 consecutive times, the interface is determined to be in working state; count the number of interfaces in all working states and take them as the number of interfaces in real-time operation; the time for the interface to collect the current value 3 times consecutively must be within a fixed time interval T.
[0055] The power consumption of outdoor emergency equipment is not a fixed value and generally adjusts dynamically with the ambient temperature. For example, in low-temperature environments, the airway heating module of the ventilator and the antifreeze module of the infusion pump will automatically start, increasing the load power consumption of the corresponding interface by 20%-30% compared to normal temperature.
[0056] Light intensity, in outdoor scenarios, directly affects the operation of additional functions of the corresponding interface load. When the light intensity is low, users need to turn on the auxiliary lighting equipment of the emergency tent. Although such equipment is not a core medical load, it will be directly included in the total energy consumption of the interface. When the light intensity is high, the lighting equipment is turned off and the additional energy consumption disappears.
[0057] Airflow velocity is negatively correlated with the power consumption of the heat dissipation module in high-power emergency equipment: when the airflow velocity is high, the natural heat dissipation effect is good, and the cooling fan only needs to run at a low speed. When the airflow velocity is low, the natural heat dissipation is insufficient, and the fan needs to run at a high speed. At this time, the energy consumption of the load corresponding to the interface is usually 2-3 times that of the high airflow velocity.
[0058] The data processing module, based on a 3D feature dataset, uses the remaining state of charge (SOC), charge / discharge rate, number of interfaces in real-time operation, ambient temperature, light intensity, and airflow speed as inputs to a Long Short-Term Memory (LSTM) neural network, and the total energy consumption of the interfaces as the output of the LSTM. This is used to train an LSTM-based model for predicting the total energy consumption of the interfaces. The specific operation is as follows:
[0059] All input and output features are normalized by min-max and mapped to the [0,1] interval. The continuous feature data are reconstructed into time series samples according to the timestamp. The time step is set to Ts, that is, the total energy consumption of the interface at the Ts+1 time is predicted based on the feature data of the first Ts time steps, forming a sample pair of input time series (Ts×6)-output label (1×1).
[0060] This invention is based on LSTM for single-step prediction. If multi-step prediction is used, the error will propagate exponentially. For example, if the prediction error of the first time is ±5%, the second time needs to be predicted based on the prediction value of the first time (including the 5% error), and the error will accumulate to ±10%-±12%. Then the error of the fifth time may exceed ±30%, which will be completely unable to support the subsequent energy supply judgment.
[0061] The time series samples were divided into training set, validation set and test set in a ratio of 7:2:1, while maintaining the continuity of time series and avoiding disrupting the time order.
[0062] The LSTM model architecture consists of four layers: an input layer, an LSTM layer, a fully connected layer, and an output layer. The specific parameters for each layer are as follows:
[0063] The input layer has a dimension of (Ts, 6), corresponding to Ts time steps and 6 input features;
[0064] The first LSTM layer has 64 hidden units, and the return sequence function is enabled (i.e., return_sequences=True). The activation function is tanh, which is used to capture shallow temporal nonlinear features.
[0065] The second LSTM layer has 32 hidden units, and the return sequence function is not enabled (i.e., return_sequences=False). The activation function is tanh, which is used to extract deep coupled temporal features.
[0066] The fully connected layer has 16 neurons and uses ReLU as the activation function to compensate for the gradient saturation problem of tanh.
[0067] The output layer has 1 neuron, corresponding to the predicted total energy consumption of the interface.
[0068] Meanwhile, initialize the remaining hyperparameters of the LSTM model: batch size of 32, learning rate of 0.001, number of iterations of 100, and dropout rate of 0.2.
[0069] The training set is input into the already built LSTM model for training. The mean squared error (MSE) is selected as the loss function. In each training iteration, the MSE loss value between the total energy consumption of the prediction interface corresponding to the input time series in the training set and the true label is calculated to quantify the model's fitting deviation on the training set.
[0070] Simultaneously, the Adam optimizer is configured to update the model parameters through backpropagation. Utilizing its adaptive learning rate characteristic, the parameters of each layer of the model are adjusted based on the MSE loss gradient of the training set, thereby improving the model's ability to fit the temporal patterns of the training set.
[0071] An early stopping mechanism is introduced during training. After each training iteration, the MSE loss of the current model parameters is evaluated immediately using the validation set. The validation set MSE is used as the monitoring indicator and a patience value of 3 is set. That is, when the validation set loss does not decrease for 3 consecutive rounds, the model training is automatically terminated to avoid the model overfitting on the training set. At the same time, the model weights when the validation set loss is the lowest are saved in real time during the training process.
[0072] The optimal model weights saved in real time during training are loaded and inference calculations are performed on the test set. The root mean square error (RMSE) and mean absolute percentage error (MAPE) are calculated respectively. RMSE directly reflects the absolute magnitude of the prediction error, while MAPE reflects the relative error level. The prediction accuracy of the model can be comprehensively evaluated by combining the two indicators.
[0073] To further improve the prediction accuracy of the obtained LSTM model, a swarm intelligence optimization algorithm can be used to optimize the time step Ts, the number of hidden units in the two-layer LSTM, and the learning rate. The swarm intelligence optimization algorithm is one of the following: mirage search optimization algorithm, RRT-based optimizer, and cloud drift optimization algorithm.
[0074] On the one hand, traditional gradient optimization algorithms rely on the gradient information of the objective function, but there is no continuous and differentiable mathematical relationship between the performance index MAPE of hyperparameters and hyperparameters, which is a typical black box optimization. On the other hand, swarm intelligence algorithms do not require gradient information and can iteratively optimize through a closed loop of three steps: parameter combination, model training, and performance evaluation.
[0075] On the other hand, traditional grid search or random search cannot efficiently handle the mixed parameter space of integers (i.e., the number of hidden units) and continuity (i.e., the learning rate), while swarm intelligence optimization algorithms can handle these two types of parameters in a unified manner through integer encoding and continuous interpolation, taking into account both the comprehensiveness and efficiency of the search.
[0076] After obtaining the optimal LSTM interface total power consumption prediction model, the energy regulation module collects the current remaining power SOC, charge and discharge rate, number of interfaces working in real time, ambient temperature, light intensity, and airflow speed, and substitutes them into the optimal LSTM interface total power consumption prediction model to obtain the total power consumption of the interface at the next moment.
[0077] The solar energy production range is calculated based on the currently collected light intensity and the time interval T between the current moment and the next moment. The specific operation is as follows:
[0078] The single-step prediction time interval of the optimal LSTM interface total energy consumption prediction model is denoted as T, which is consistent with the above fixed time interval. The current light intensity is denoted as L0, and the current airflow speed is denoted as v.
[0079] Based on the inherent parameters of solar panels, namely conversion efficiency α and panel area S, according to formula P base =L0×ɳ×S÷10 6 To calculate the basic output power, taking into account the natural fluctuations in light intensity over a short period, a fluctuation coefficient k is set, and the upper limit P of the basic production capacity is calculated. up =P base ×(1+k), lower limit P low =P base ×(1-k), we obtain the basic range of solar energy production capacity as [P low ×T÷60,P up [×T÷60];
[0080] The corrected solar panel conversion efficiency ɳ' is calculated using the formula ɳ'=ɳ+(v-v0)×△ɳ, where v o The optimal airflow speed for solar panels is the standard airflow velocity, at which the solar panels dissipate heat best and achieve the highest conversion efficiency; △ɳ represents the change in the solar panel's conversion efficiency, and both need to be preset.
[0081] Substitute ɳ' into P base =L0×ɳ×S÷10 6 Calculate P up '、P low ', thus obtaining the final solar energy production capacity range E solar =[P low '×T÷60,P up [×T÷60];
[0082] The total energy consumption of the interface at the next time step, obtained from the optimal LSTM interface total energy consumption prediction model, is compared with the solar energy production range to classify the scenarios. Based on the classified scenarios, different energy regulation strategies are adopted, including:
[0083] If it is Scenario 1, the following energy regulation strategy will be implemented: The functional operation status of all interfaces and their corresponding loads will be collected in real time through the unified communication protocol to confirm that the core loads are in normal working mode; at the same time, for non-core loads, the corresponding real-time power benchmark value will be derived based on the preset energy consumption benchmark value within T time period, that is, power benchmark value = energy consumption benchmark value ÷ T. The current operating power of non-core loads will be locked at this benchmark value, and they will be prohibited from increasing their power on their own. This will ensure that the energy consumption of non-core loads within T time period does not exceed the preset benchmark and avoid occupying the energy quota of core loads.
[0084] The core load refers to the load equipment that is directly related to the patient's life safety, supports the core diagnosis and treatment actions of outdoor emergency rescue, and cannot be temporarily interrupted or have its power reduced. Its operating status directly determines the emergency rescue effect. Even if the power supply is tight, it is necessary to give priority to ensuring its full power and uninterrupted operation.
[0085] The non-core loads refer to load devices that provide auxiliary convenience for outdoor emergency rescue, do not directly participate in core diagnosis and treatment actions, and will not threaten the patient's life safety after interruption or power reduction. In the case of energy shortage, energy can be saved by limiting power and suspending operation, and priority should be given to ensuring the power supply of core loads.
[0086] If it is scenario two, the following energy regulation strategy will be implemented: Based on the total energy consumption benchmark of the interface within the current time T, the corresponding real-time total power benchmark will be derived; the total power that can be reduced for all non-core loads will be calculated in real time (this power reduction corresponds to the energy reduction within time T, i.e., energy reduction = power reduction × T), ensuring that its proportion of the current real-time total power of the interface is ≤15%, so as to avoid excessive reduction of non-core loads leading to failure of auxiliary functions.
[0087] At the same time, the unified communication protocol is used to provide real-time feedback on the adjusted status, focusing on confirming that the operating power of the core load does not fluctuate, ensuring that the energy supply of the core load is stable within time T, and that there is no energy gap due to the reduction of non-core loads.
[0088] If it is scenario three, then the following energy regulation strategy will be implemented: based on the current available energy range E available Predicted energy consumption E for the next T cycle pred Derivation of the core load priority energy consumption quota = E within time T. available The lower limit is multiplied by 80% and converted into a real-time power reference to ensure that more than 80% of the battery's output energy is preferentially supplied to the core load through this power reference.
[0089] If the core load experiences a momentary high power demand (corresponding to the peak of momentary energy consumption demand), it will automatically draw on the energy consumption reduction redundancy of the non-core load to supplement it. That is, the redundancy power = the power that can be reduced by the non-core load = the corresponding energy consumption reduction amount ÷ T, where the energy consumption reduction amount is the difference between the baseline energy consumption and the actual energy consumption of the non-core load.
[0090] Real-time monitoring of the total power reduction ratio of non-core loads ensures that the energy consumption reduction ratio within the corresponding time period T is ≤20%, in order to avoid excessive reduction leading to failure of auxiliary functions; at the same time, emergency power replenishment prompts are pushed through the human-machine interface of the portable power supply, guiding users to connect backup lithium battery modules or adjust the orientation of solar panels to increase production capacity, until the matching degree between the available energy range and the predicted energy consumption returns to the range of scenario one or scenario two, ensuring that the continuous energy supply of core loads is not affected.
[0091] Example 2
[0092] This embodiment further discloses a method for scene marking based on embodiment 1, such as... Figure 2 As shown, the specific content includes:
[0093] The battery's remaining state of charge (SOC) is converted into actual remaining energy (E) using an extended Kalman filter algorithm. bat The specific formula is E bat =Q×SOC, and then compared with the solar energy capacity range E. solar Combining these, we obtain the fundamental energy range E. base =[E solar_low +E bat E solar_high +E bat Where Q is the rated capacity of the battery, and E is the rated capacity of the battery. solar_low E solar_high Solar energy capacity range E solar The lower and upper limits;
[0094] Collect the current battery charge / discharge rate C rate If the device is in a charging state, calculate the additional charging energy E within time T. extra =C rate ×Q×T÷60, E extra Superimposed on E available The upper and lower limits are used to obtain the final usable energy range E. available =[E base_low +E extra E extra_high +E extra ], where E base_low E extra_high Basic energy range E base The lower and upper limits;
[0095] Call the trained optimal LSTM interface total energy consumption prediction model and output the predicted interface total energy consumption value E at the next time point (i.e., after time T). pred Simultaneously, based on the error distribution during the model training phase, a prediction confidence interval E with a 95% confidence level is output.pred_conf =[E pred -△E,E pred +△E], where △E is the critical error value;
[0096] The method for determining the value of △E is as follows: Calculate the single-step prediction error of the optimal LSTM interface total energy consumption prediction model for all samples on the test set. Since the error of the time series prediction model usually follows a normal distribution, that is, the error is concentrated near the mean μ of the single-step prediction error of the group, the dispersion of the error can be represented by the standard deviation σ. The statistical law of the normal distribution is that about 95% of the error will fall within the range of mean ± 1.96 times the standard deviation. Therefore, we get △E = 1.96 × σ + |μ|.
[0097] Determine E pred_conf With E available The overlapping interval [E] overlap_min E overlap_max ], where E overlap_min =max(E pred -△E,E base_low +E extra E overlap_max =min(E pred +△E,E extra_high +E extra );
[0098] If E overlap_min ≤E overlap_max Then calculate the overlap R=(E overlap_max -E overlap_min ) / (E pred +△E-(E pred -△E)); If E overlap_min >E overlap_max If R = 0, then R = 0;
[0099] The overlap R quantifies the degree of matching between the reliable range of the predicted total power consumption of the interface and the available energy range. The larger the value, the higher the degree of matching between supply and demand.
[0100] Calculate E available The midpoint value E available_mid The predicted total power consumption E of the interface is obtained. pred With midpoint value E available_mid The deviation ratio D = |E pred -E available_mid | / E available_mid ;
[0101] The deviation ratio D quantifies the degree of deviation between the predicted total power consumption of the interface and the available energy center. The smaller the value, the more sufficient the redundancy of the energy supply.
[0102] Scene labeling is based on R and D, including: if R≥R1 and D≤D1, it is labeled as Scene 1; if R2≤R<R1 or D1<D≤D2, it is labeled as Scene 2; if R<R2 or D>D2, it is labeled as Scene 3. Here, R1 and R2 are the lower and upper limits of the overlap R, and D1 and D2 are the lower and upper limits of the deviation ratio D, which need to be preset according to the actual situation.
[0103] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.
[0104] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. An intelligent management system for a portable outdoor energy storage power supply, characterized in that, include: The data acquisition module collects the remaining SOC and charge / discharge rate of the battery pack from the battery status dimension, the total energy consumption of the interface and the number of interfaces working in real time from the interface status dimension, and the ambient temperature, light intensity and airflow speed from the environmental status dimension at fixed time intervals of T minutes. The three types of data are aligned using timestamp synchronization technology to obtain a three-dimensional feature dataset. The data processing module, based on a three-dimensional feature dataset, uses the remaining power SOC, charge and discharge rate, number of interfaces working in real time, ambient temperature, light intensity, and airflow speed as inputs to the LSTM and the total energy consumption of the interfaces as the output of the LSTM. After training, the optimal LSTM total energy consumption prediction model for the interfaces is obtained. The energy regulation module collects the current remaining power SOC, charge / discharge rate, number of interfaces in real time, ambient temperature, light intensity, and airflow speed, and substitutes them into the optimal LSTM interface total energy consumption prediction model to predict the total power consumption of the interface at the next moment. At the same time, it calculates the solar energy production range based on the currently collected light intensity and the time interval T between the current moment and the next moment, and compares the predicted total power consumption of the interface at the next moment with the solar energy production range to mark the scene, thereby adopting different energy regulation strategies.
2. The intelligent management system for outdoor portable energy storage power supply according to claim 1, characterized in that, The specific method for collecting the number of interfaces in real-time operation is as follows: assign a unique physical identifier to each interface, and detect whether there is current output in real time through the miniature current sensor built into the interface; if the current value of an interface is greater than 0 for three consecutive times, the interface is determined to be in working state; count the number of interfaces in all working states and take them as the number of interfaces in real-time operation; the time for the interface to collect the current value three times consecutively must be within a fixed time interval T.
3. The intelligent management system for an outdoor portable energy storage power supply according to claim 1, characterized in that, The specific steps for training the LSTM interface total energy consumption prediction model are as follows: All input and output features are normalized by min-max and mapped to the [0,1] interval. At the same time, continuous feature data are reconstructed into time series samples according to timestamps. The time step is set to Ts to form input time series sequence-output label sample pairs. The time series samples are divided into training set, validation set and test set in a ratio of 7:2:
1. Build a four-layer architecture for the LSTM model, consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer, and set the corresponding parameters. The training set is substituted into the already built LSTM model for training, and the mean squared error (MSE) is selected as the loss function. At the same time, the Adam optimizer is configured to perform backpropagation updates of the model parameters. After each training iteration, the MSE loss of the current model parameters is immediately evaluated using the validation set. The validation set MSE is used as the monitoring indicator and the patience value is set to 3. That is, when the validation set loss does not decrease for 3 consecutive rounds, the model training is automatically terminated. At the same time, the model weights with the lowest validation set loss during the training process are saved and marked as the optimal model weights. Load the optimal model weights saved during training to obtain the optimal LSTM interface total energy consumption prediction model, and perform inference calculations on the test set to calculate the root mean square error (RMSE) and mean absolute percentage error (MAPE).
4. The intelligent management system for an outdoor portable energy storage power supply according to claim 3, characterized in that, The specific parameters for the four layers are as follows: the dimension of the input layer is (Ts, 6); the number of hidden units in the first LSTM layer is 64, the return sequence function is enabled, and the activation function is tanh; the number of hidden units in the second LSTM layer is 32, the return sequence function is not enabled, and the activation function is tanh; the number of neurons in the fully connected layer is 16, and the activation function is ReLU; the number of neurons in the output layer is 1; the batch size is 32, the learning rate is 0.001, the number of iterations is 100, and the dropout rate is 0.
2.
5. The intelligent management system for an outdoor portable energy storage power supply according to claim 3, characterized in that, A swarm intelligence optimization algorithm is used to optimize the time step Ts, the number of hidden units in two LSTM layers, and the learning rate of the LSTM model. The swarm intelligence optimization algorithm is one of the following: mirage search optimization algorithm, RRT-based optimizer, and cloud drift optimization algorithm.
6. The intelligent management system for an outdoor portable energy storage power supply according to claim 1, characterized in that, The specific steps for calculating the solar energy production range are as follows: The single-step prediction time interval of the optimal LSTM interface total energy consumption prediction model is denoted as T, which is consistent with the fixed time interval. The current light intensity is denoted as L0, and the current airflow speed is denoted as v. Based on the solar panel's conversion efficiency α and panel area S, according to formula P base =L0×ɳ×S÷10 6 Calculate the basic output power P base Set a fluctuation coefficient k and calculate the upper limit P of the basic production capacity. up =P base ×(1+k), lower limit P low =P base ×(1-k), we obtain the basic range of solar energy production capacity as [P low ×T÷60,P up [×T÷60]; The corrected solar panel conversion efficiency ɳ' is calculated using the formula ɳ'=ɳ+(v-v0)×△ɳ, where v o The standard airflow velocity is the most suitable for solar panels, and Δɳ is the change in the conversion efficiency of the solar panels. Substitute ɳ' into P base =L0×ɳ×S÷10 6 Calculate P up '、P low ', thus obtaining the final solar energy production capacity range E solar =[P low '×T÷60,P up '×T÷60'.
7. The intelligent management system for an outdoor portable energy storage power supply according to claim 1, characterized in that, The specific steps for scene marking are as follows: The battery's remaining state of charge (SOC) is converted into actual remaining energy (E) using an extended Kalman filter algorithm. bat The specific formula is E bat =Q×SOC, and then compared with the solar energy capacity range E. solar Combining these, we obtain the fundamental energy range E. base =[E solar_low +E bat E solar_high +E bat Where Q is the rated capacity of the battery, and E is the rated capacity of the battery. solar_low E solar_high Solar energy capacity range E solar The lower and upper limits; Collect the current battery charge / discharge rate C rate If the device is in a charging state, calculate the additional charging energy E within time T. extra =C rate ×Q×T÷60, E extra Superimposed on E available The upper and lower limits are used to obtain the final usable energy range E. available =[E base_low +E extra E extra_high +E extra ], where E base_low E extra_high Basic energy range E base The lower and upper limits; Invoke the trained optimal LSTM interface total energy consumption prediction model and output the predicted interface total energy consumption value E for the next time point. pred Based on the model's error distribution on the test set, it outputs a 95% confidence interval E for prediction. pred_conf =[E pred -△E,E pred +△E], where △E is the critical error value; Determine E pred_conf With E available The overlapping interval [E] overlap_min E overlap_max ], where E overlap_min =max(E pred -△E,E base_low +E extra E overlap_max =min(E pred +△E,E extra_high +E extra ); If E overlap_min ≤E overlap_max Then calculate the overlap R=(E overlap_max -E overlap_min ) / (E pred +△E-(E pred -△E)); If E overlap_min >E overlap_max If R = 0, then R = 0; Calculate E available The midpoint value E available_mid The predicted total power consumption E of the interface is obtained. pred With midpoint value E available_mid The deviation ratio D = |E pred -E available_mid | / E available_mid ; The scene is labeled according to R and D: if R≥R1 and D≤D1, it is labeled as scene one; if R2≤R<R1 or D1<D≤D2, it is labeled as scene two; if R<R2 or D>D2, it is labeled as scene three. Here, R1 and R2 are the lower and upper limits of the overlap R, and D1 and D2 are the lower and upper limits of the deviation ratio D.
8. The intelligent management system for an outdoor portable energy storage power supply according to claim 7, characterized in that, The method for determining the value of △E is as follows: calculate the single-step prediction error of the optimal LSTM interface total energy consumption prediction model on all samples of the test set, obtain the average value μ and standard deviation σ of the data set, and calculate the critical error value △E according to the formula △E=1.96×σ+|μ|.
9. The intelligent management system for an outdoor portable energy storage power supply according to claim 1, characterized in that, Different energy regulation strategies based on labeling scenarios specifically include: If it is Scenario 1, the following energy regulation strategy will be implemented: The functional operation status of the load corresponding to all interfaces will be collected in real time through the unified communication protocol to confirm that the core loads are in normal working mode; at the same time, for non-core loads, the corresponding real-time power benchmark value will be derived based on the preset energy consumption benchmark value within T time period, that is, power benchmark value = energy consumption benchmark value ÷ T, and the current operating power of non-core loads will be locked at this benchmark value to prevent them from increasing their power on their own. If it is scenario two, the following energy regulation strategy will be implemented: Based on the total energy consumption benchmark of the interface within the current time T, the corresponding real-time total power benchmark will be derived; the total power that can be reduced for all non-core loads will be counted in real time to ensure that its proportion of the current real-time total power of the interface is ≤15%; at the same time, the adjusted status will be fed back in real time through the unified communication protocol to confirm that the operating power of the core loads is stable. If it is scenario three, then the following energy regulation strategy will be implemented: based on the current available energy range E available Predicted energy consumption E for the next T cycle pred Derivation of the core load priority energy consumption quota = E within time T. available The lower limit is multiplied by 80% and converted into a real-time power benchmark to ensure that more than 80% of the battery's output energy is supplied to the core load through this power benchmark. At the same time, emergency power replenishment prompts are pushed through the human-machine interface of the portable power supply to guide users to connect backup lithium battery modules or adjust the orientation of solar panels to increase production capacity until the matching degree between the available energy range and the predicted energy consumption returns to the range of Scenario 1 or Scenario 2.
10. The intelligent management system for an outdoor portable energy storage power supply according to claim 9, characterized in that, The core load refers to load devices that are directly related to the patient's life safety, support the core diagnosis and treatment actions of outdoor emergency care, and cannot be temporarily interrupted or have their power reduced; the non-core load refers to load devices that provide auxiliary convenience for outdoor emergency care, do not directly participate in the core diagnosis and treatment actions, and will not threaten the patient's life safety if their power is interrupted or reduced.