Self-maintenance system based on monitoring of operating state difference of single storage battery

Abstract

The invention provides a self-maintenance system based on single storage battery operation state difference monitoring, which comprises a distributed monitoring module array, each monitoring unit comprises a multi-parameter fusion sensor group, and the distributed monitoring module array is used for acquiring the voltage fluctuation frequency spectrum, temperature gradient distribution, internal resistance change track and electrolyte ion migration rate of a single battery in real time; and the dynamic difference evaluation module is used for processing the multi-dimensional time sequence data by adopting a time sequence convolutional neural network and calculating the operating state difference degree among the monomers. Through the distributed monitoring module array and the multi-parameter fusion sensor group, multi-dimensional data such as the voltage, the temperature, the internal resistance and the electrolyte ion migration rate of the single batteries are collected in real time, and the operation state difference between the single batteries can be accurately quantified in combination with the time sequence convolutional neural network and the dynamic difference evaluation algorithm. Abnormal monomers are effectively identified, and a reliable basis is provided for a subsequent maintenance strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of battery monitoring, and particularly to a self-maintenance system based on the monitoring of the operating state differences of individual storage batteries. Background Art

[0002] With the rapid development of fields such as electric vehicles and energy storage systems, as the core energy storage unit, the performance and lifespan of storage batteries directly affect the reliability and economy of the system. However, in actual operation of storage batteries, due to manufacturing differences between monomers, inconsistencies in the usage environment, and differences in the aging rate, significant differences in the operating states of individual batteries often occur. Such differences not only reduce the overall performance of the battery pack but may also pose safety hazards such as thermal runaway. Traditional battery management systems usually adopt a passive maintenance strategy based on voltage equalization, which is difficult to effectively cope with complex operating state differences and lacks the ability to accurately predict the health state of the battery and perform adaptive maintenance.

[0003] Therefore, in response to the above problems, a self-maintenance system based on the monitoring of the operating state differences of individual storage batteries is now being developed. Summary of the Invention

[0004] In order to overcome the drawbacks of the prior art, the present invention provides a self-maintenance system based on the monitoring of the operating state differences of individual storage batteries.

[0005] The technical solution of the present invention is: A self-maintenance system based on the monitoring of the operating state differences of individual storage batteries, comprising:

[0006] A distributed monitoring module array, each monitoring unit includes a multi-parameter fusion sensor group, which real-time collects the voltage fluctuation spectrum, temperature gradient distribution, internal resistance change trajectory, and electrolyte ion migration rate of individual batteries;

[0007] A dynamic difference evaluation module, which uses a time series convolutional neural network to process multi-dimensional time series data and calculates the operating state difference degree between each monomer. The difference degree calculation includes: constructing a three-dimensional state space to map the co-variation pattern of voltage-temperature-internal resistance, quantifying the morphological differences of the operating trajectories between different monomers through the dynamic time warping algorithm, and introducing a dispersion evaluation index based on information entropy to characterize the group consistency level;

[0008] An adaptive maintenance strategy generator, which establishes a two-layer decision-making model according to the real-time difference degree: the first layer divides the difference level interval based on a fuzzy inference system, and the second layer uses a reinforcement learning algorithm to perform dynamic game between maintenance cost and performance gain;

[0009] A heterogeneous actuator cluster, which includes an active equalizer group, a phase change temperature control unit, and an electrolyte microcirculation device, and can perform three types of maintenance operations: difference compensation, thermal management optimization, and chemical environment repair.

[0010] As a preferred embodiment of the present invention, the multi-parameter fusion sensor group is integrated with: a high-frequency impedance spectrum analysis unit, a distributed optical fiber temperature measurement network, a micro Hall array sensor, and a flexible printed circuit type voltage acquisition matrix. Feature extraction is performed on the data of each sensor through a time-frequency domain joint analysis method, specifically including:

[0011] Step 1: Perform wavelet packet decomposition on the voltage signal to obtain the energy ratio of 16 sub-bands;

[0012] Step 2: Use empirical mode decomposition to process the temperature sequence to extract the intrinsic mode function;

[0013] Step 3: Establish an equivalent circuit model parameter set through complex impedance plane fitting.

[0014] As a preferred embodiment of the present invention, the dynamic difference evaluation module includes:

[0015] A state encoder that converts multi-source monitoring data into a 128-dimensional feature vector;

[0016] A spatio-temporal attention mechanism that is used to calculate the feature weights at different time steps and spatial positions;

[0017] A difference degree calculation unit that uses the Wasserstein distance to measure the state distribution difference in the difference between monomers, constructs an eigenvalue spectrum analysis model based on the covariance matrix for the group dispersion, and performs dynamic threshold adjustment during the process. According to the operating stage of the battery pack, namely the three stages of charging, discharging, and standing, the determination benchmark is automatically corrected, and the output result is a composite difference index including a short-term fluctuation index and a long-term degradation trend.

[0018] As a preferred embodiment of the present invention, the working process of the adaptive maintenance strategy generator includes:

[0019] Step S1: Establish a three-dimensional decision space coordinate system, namely the difference degree, maintenance cost, and expected benefit;

[0020] Step S2: Use Monte Carlo tree search to generate a candidate strategy set;

[0021] Step S3: Evaluate the Q-value function of each strategy through a value network;

[0022] Step S4: Introduce an environmental perception compensation mechanism to dynamically adjust the strategy weights according to external factors such as the current environmental temperature and vibration intensity;

[0023] Step S5: Output the optimal maintenance instruction sequence and the execution timing plan.

[0024] As a preferred embodiment of the present invention, it further includes a battery pack life prediction method based on multi-modal data fusion, and the specific steps are as follows:

[0025] Step 1: Construct a cross-domain feature matrix including electrical, thermodynamic, and chemical parameters;

[0026] Step 2: Train a spatio-temporal graph convolutional network to capture the coupling effect between cells;

[0027] Step 3: Establish an attention-guided long short-term memory model to predict the remaining life;

[0028] Step 4: Feed the prediction result back to the maintenance strategy generator to dynamically optimize the maintenance cycle parameters;

[0029] Step 5: Verify the effectiveness of the prediction model through a digital twin system and update the parameters online.

[0030] As a preferred embodiment of the present invention, the spatio-temporal graph convolutional network in Step 2 includes:

[0031] Spatial graph convolutional layer: Define the battery connection topology graph as graph structure data;

[0032] Temporal convolutional layer: Use dilated convolution to capture long-term temporal dependencies;

[0033] Dynamic edge weight calculation module: Update the connection weights between nodes according to real-time operation data;

[0034] Cross-domain feature fusion gate: Integrate the information flow of electrical features and thermodynamic features.

[0035] As a preferred embodiment of the present invention, it further includes a self-healing communication network, and the self-healing communication network has a redundant channel dynamic switching mechanism, a hybrid communication protocol based on Lora and CAN bus, a data integrity verification module, and an electromagnetic interference shielding structure.

[0036] As a preferred embodiment of the present invention, a pulsed energy injection technology is introduced in the maintenance execution stage, which specifically includes a high-frequency resonant equalization circuit, a controllable avalanche energy release device, a nanosecond pulse generator based on GaN devices, and an energy recovery unit.

[0037] By adopting the above technical solutions, the present invention has the following advantages:

[0038] 1. Through the distributed monitoring module array and multi-parameter fusion sensor group, the present invention can collect multi-dimensional data such as the voltage, temperature, internal resistance, and electrolyte ion migration rate of individual batteries in real time. Combining with the time series convolutional neural network and the dynamic difference evaluation algorithm, it can accurately quantify the operation state differences between individual cells, effectively identify abnormal cells, and provide a reliable basis for subsequent maintenance strategies.

[0039] 2. The present invention adopts an adaptive maintenance strategy generator, which dynamically optimizes the maintenance strategy based on the Monte Carlo tree search and reinforcement learning algorithms. Combining the pulsed energy injection technology and the digital twin system, it significantly improves the maintenance efficiency, extends the service life of the battery pack, and reduces the maintenance cost and safety risks at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic structural diagram of the present invention.

[0041] Figure 2 It is a schematic structural diagram of the multi-parameter fusion sensor group of the present invention.

[0042] Figure 3 It is a flowchart of the steps for feature extraction by the time-frequency domain joint analysis method of the present invention.

[0043] Figure 4 It is a schematic structural diagram of the dynamic difference evaluation module of the present invention.

[0044] Figure 5 It is a flowchart of the working steps of the adaptive maintenance strategy generator of the present invention.

[0045] Figure 6 It is a flowchart of the steps of the battery pack life prediction method based on multi-modal data fusion of the present invention.

[0046] Figure 7 It is a schematic structural diagram of the spatio-temporal graph convolutional network of the present invention.

[0047] Figure 8 It is a schematic structural diagram of the pulsed energy injection technology of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] Referring to the embodiments herein means that the specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the present invention. The phrase appears in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0049] The self-maintenance system based on the monitoring of the operating state differences of individual storage batteries, such as Figure 1As shown in the figure, it includes a distributed monitoring module array, a dynamic difference evaluation module, an adaptive maintenance strategy generator, a heterogeneous actuator cluster, and a self-healing communication network. In the distributed monitoring module array, each monitoring unit contains a multi-parameter fusion sensor group that can collect the voltage fluctuation spectrum, temperature gradient distribution, internal resistance change trajectory, and electrolyte ion migration rate of a single battery in real time. The dynamic difference evaluation module uses a time series convolutional neural network to process multi-dimensional time series data and calculate the operation state difference degree between each single battery. The difference degree calculation includes: constructing a three-dimensional state space to map the co-variation pattern of voltage-temperature-internal resistance, quantifying the morphological difference of the operation trajectories between different single batteries through the dynamic time warping algorithm, and introducing a dispersion evaluation index based on information entropy to characterize the group consistency level. The adaptive maintenance strategy generator will establish a two-layer decision-making model based on the real-time difference degree. Specifically, the first layer divides the difference level interval based on a fuzzy inference system, and the second layer uses a reinforcement learning algorithm to perform dynamic game between maintenance cost and performance gain. The heterogeneous actuator cluster includes an active balancer group, a phase change temperature control unit, and an electrolyte microcirculation device, which can perform three types of maintenance operations: difference compensation, thermal management optimization, and chemical environment repair. The self-healing communication network has a redundant channel dynamic switching mechanism, a hybrid communication protocol based on Lora and CAN bus, a data integrity verification module, and an electromagnetic interference shielding structure.

[0050] As Figure 2 shown, the multi-parameter fusion sensor group is integrated with: a high-frequency impedance spectrum analysis unit, a distributed optical fiber temperature measurement network, a micro Hall array sensor, and a flexible printed circuit type voltage acquisition matrix. The data of each sensor is subjected to feature extraction through a time-frequency domain joint analysis method. As Figure 3 shown, it specifically includes:

[0051] Step 1: By performing wavelet packet decomposition on the voltage signal, the signal is converted from the time domain to the frequency domain, and the energy characteristics of different frequency bands are extracted. Specifically, during implementation, first select an appropriate wavelet basis function, perform 4-layer decomposition on the voltage signal to obtain 16 sub-frequency bands. Each sub-frequency band corresponds to a different frequency range and can reflect the high-frequency noise, medium-frequency fluctuation, and low-frequency trend components in the voltage signal. Subsequently, calculate the energy value of each sub-frequency band and obtain the energy proportion of each frequency band through normalization processing;

[0052] Step 2: Using the empirical mode decomposition method, decompose the temperature sequence into a series of intrinsic mode functions. Each IMF represents the temperature change characteristics at different time scales. Specifically, during implementation, first identify the extreme points of the temperature sequence, generate the upper and lower envelope lines through interpolation, calculate the mean curve, and extract the IMF through iteration until the remaining signal meets the monotonicity or extreme point quantity requirements. After decomposition, the high-frequency IMF can reflect local temperature mutations, while the low-frequency IMF characterizes the overall temperature rise trend;

[0053] Step 3: Establish an equivalent circuit model of the battery through complex impedance plane fitting, extract key parameters to evaluate the health status of the battery. Specifically, during implementation, first measure the complex impedance values of the battery at multiple frequency points and plot the Nyquist diagram in the complex plane. Subsequently, select the Randles model and use the nonlinear least squares method to fit the measured data to obtain model parameters (such as ohmic resistance, charge transfer resistance, double-layer capacitance, etc.).

[0054] As Figure 4 shown, the dynamic difference evaluation module includes:

[0055] A state encoder that converts multi-source monitoring data into a 128-dimensional feature vector;

[0056] A spatio-temporal attention mechanism used to calculate the feature weights at different time steps and spatial positions;

[0057] A difference degree calculation unit that uses the Wasserstein distance to measure the state distribution difference in the differences between monomers, constructs an eigenvalue spectrum analysis model based on the covariance matrix for the group dispersion degree, and dynamically adjusts the threshold during the process. According to the operating stages of the battery pack, namely the charging, discharging, and static stages, it automatically corrects the judgment benchmark and outputs the result as a composite difference index including a short-term fluctuation index and a long-term degradation trend.

[0058] As Figure 5 shown, the working process of the adaptive maintenance strategy generator includes:

[0059] Step S1: Provide a quantitative framework for formulating maintenance strategies by constructing a three-dimensional decision space of difference degree, maintenance cost, and expected benefit. The difference degree is measured by the dynamic difference degree index, which reflects the inconsistency of the operating states between monomers in the battery pack. The maintenance cost includes a comprehensive evaluation of energy consumption, time cost, and hardware loss. The expected benefit is quantified by indicators such as the improvement in health status, enhancement of safety performance, and extension of lifespan;

[0060] Step S2: Use the Monte Carlo tree search algorithm to efficiently search for candidate maintenance strategies in the decision space. Specifically, during implementation, first initialize the root node of the search tree, representing the current system state. Subsequently, through the four stages of selection, expansion, simulation, and backtracking, gradually explore possible maintenance actions (such as active balancing, temperature control optimization, chemical repair, etc.). In the simulation stage, evaluate the short-term effects of the strategies through random sampling. In the backtracking stage, update the visit times and cumulative benefit values of the nodes, and finally generate a set of candidate strategy sets, covering a variety of options from low-cost conservative strategies to high-cost aggressive strategies;

[0061] Step S3: Evaluate the Q-value function of each strategy through the value network;

[0062] Step S4: Introduce an environmental perception compensation mechanism to dynamically adjust the policy weights according to external factors such as the current environmental temperature and vibration intensity;

[0063] Step S5: Generate an optimal maintenance instruction sequence and plan the execution timing according to the evaluation results. Specifically, during implementation, first decompose the policy with the highest Q value into specific maintenance actions (such as the magnitude of the balanced current, the temperature control target value, the chemical repair dosage, etc.), and arrange the execution order according to the dependency relationship between the actions. Subsequently, according to the real-time state of the system and resource constraints, formulate an execution timing plan to ensure that the maintenance process is efficient and does not affect the normal operation of the system. The finally output instruction sequence includes the action type, parameter settings, execution time, and expected effects, providing clear guidance for the maintenance operation.

[0064] Specifically, in step S3, the formula for the Q-value function is:

[0065] Q(s,a) = α·ΔSOH - β·E cost + γ·P safe ;

[0066] Where α, β, and γ are weight coefficients used to balance the influence of different factors on the overall evaluation. ΔSOH represents the change in the health state, reflecting the increase or decrease in the system health state after taking an action. E cost represents the energy cost required to perform a certain action, and P safe is a metric representing safety, which involves the safety limit of the operation or risk assessment.

[0067] It also includes a battery pack life prediction method based on multi-modal data fusion. As Figure 6 shown, the specific steps are as follows:

[0068] Step 1: By integrating multi-source monitoring data, construct a cross-domain feature matrix to provide comprehensive input features for life prediction. Specifically, during implementation, extract temporal features such as voltage, current, and internal resistance from the electrical domain, extract spatial features such as temperature distribution, temperature rise rate, and heat flux density from the thermodynamics domain, and extract chemical characteristic parameters such as electrolyte concentration, SEI film thickness, and ion migration rate from the chemical domain. Align these features according to time steps to form a multi-dimensional feature matrix. Each row of the matrix corresponds to a time point, and each column corresponds to a feature dimension. Through standardization and normalization processing, eliminate the influence of different feature dimensions and provide high-quality input data for subsequent model training;

[0069] Step 2: Use the spatio-temporal graph convolutional network to model the coupling effect between individual cells in the battery pack. Specifically, during implementation, first abstract the battery pack topology into graph-structured data, where nodes represent individual cells and edges represent electrical connection relationships. Subsequently, extract the spatial correlation features between individual cells through the spatial graph convolutional layer, and capture the evolution law in the time dimension through the temporal convolutional layer. During the training process, use the mean squared error as the loss function, and optimize the network parameters through backpropagation. Eventually, the model can accurately capture the mutual influence between individual cells, providing more accurate input features for life prediction;

[0070] Step 3: Predict the remaining useful life of the battery pack through the attention-guided long short-term memory model. Specifically, during implementation, first input the spatio-temporal features extracted by ST-GCN into the LSTM network to capture the long-term dependencies in the time series. Subsequently, introduce the attention mechanism to dynamically allocate weights to features at different time steps, focusing on features that have a significant impact on life attenuation (such as sudden drops in capacity, sudden increases in internal resistance, etc.). The output of the model is the remaining cycle times or remaining available capacity of the battery pack, and the uncertainty of the prediction result is evaluated through the confidence interval;

[0071] Step 4: Dynamically optimize the maintenance cycle by feeding the life prediction result back to the maintenance strategy generator. Specifically, during implementation, first calculate the health state index based on the predicted remaining life and evaluate the effectiveness of the current maintenance strategy. Subsequently, adjust the maintenance cycle parameters (such as equalization interval, temperature control threshold, chemical repair frequency, etc.) through an optimization algorithm to make the maintenance strategy match the actual attenuation state of the battery pack. For example, when the predicted remaining life is short, shorten the maintenance interval to delay attenuation; when the predicted remaining life is long, appropriately extend the maintenance interval to reduce maintenance costs;

[0072] Step 5: Use the digital twin system to verify and optimize the life prediction model. Specifically, during implementation, first build a virtual model of the battery pack on the digital twin platform to simulate the actual operating environment and working conditions. Subsequently, compare the output of the prediction model with the simulation results of the digital twin system to evaluate the prediction accuracy and robustness of the model. Update the model parameters through an online learning algorithm according to the comparison results, enabling it to adapt to the aging characteristics and changes in the operating environment of the battery pack.

[0073] Specifically, as Figure 7 shown, the spatio-temporal graph convolutional network in Step 2 includes a spatial graph convolutional layer, which defines the battery connection topology graph as graph-structured data; a temporal convolutional layer, which uses dilated convolution to capture long temporal dependencies; a dynamic edge weight calculation module, which updates the connection weights between nodes according to real-time operating data; and a cross-domain feature fusion gate, which integrates the information flows of electrical features and thermodynamic features.

[0074] Specifically, the pulse energy injection technology is introduced during the maintenance execution stage, such as Figure 8As shown, it specifically includes a high-frequency resonance equalization circuit, a controllable avalanche energy release device, a nanosecond pulse generator based on GaN devices, and an energy recovery unit.

[0075] The above embodiments are provided for those skilled in the art to implement or use the present invention. Those skilled in the art can make various modifications or changes to the above embodiments without departing from the idea of the present invention. Therefore, the protection scope of the present invention is not limited by the above embodiments, but should be the maximum scope that conforms to the innovative features mentioned in the claims.

Claims

1. A self-maintenance system based on the monitoring of the operating state differences of individual storage batteries, characterized in that It includes: A distributed monitoring module array, where each monitoring unit contains a multi-parameter fusion sensor group that can collect the voltage fluctuation spectrum, temperature gradient distribution, internal resistance change trajectory, and electrolyte ion migration rate of individual batteries in real time. A dynamic difference evaluation module that uses a time series convolutional neural network to process multi-dimensional time series data and calculate the operating state difference degree between individual cells. The difference degree calculation includes: constructing a three-dimensional state space to map the co-variation pattern of voltage-temperature-internal resistance, quantifying the morphological differences of operating trajectories between different cells through the dynamic time warping algorithm, and introducing a dispersion evaluation index based on information entropy to characterize the group consistency level. An adaptive maintenance strategy generator that establishes a two-layer decision-making model based on the real-time difference degree: the first layer divides the difference level interval based on a fuzzy inference system, and the second layer uses a reinforcement learning algorithm to perform dynamic games between maintenance costs and performance benefits. A heterogeneous actuator cluster that includes an active balancer group, a phase change temperature control unit, and an electrolyte microcirculation device, which can perform three types of maintenance operations: difference compensation, thermal management optimization, and chemical environment repair.

2. The self-maintenance system based on the monitoring of the operation state differences of individual storage batteries according to claim 1, wherein, The multi-parameter fusion sensor group is integrated with: a high-frequency impedance spectrum analysis unit, a distributed optical fiber temperature measurement network, a micro Hall array sensor, and a flexible printed circuit type voltage acquisition matrix. Feature extraction is performed on the data of each sensor through a time-frequency domain joint analysis method, specifically including: Step 1: Perform wavelet packet decomposition on the voltage signal to obtain the energy ratio of 16 sub-bands. Step 2: Use empirical mode decomposition to process the temperature sequence and extract the intrinsic mode functions. Step 3: Establish an equivalent circuit model parameter set through complex impedance plane fitting.

3. The self-maintenance system based on the monitoring of the operating state differences of individual storage batteries according to claim 1, characterized in that The dynamic difference evaluation module includes: A state encoder that converts multi-source monitoring data into a 128-dimensional feature vector. A spatio-temporal attention mechanism that is used to calculate the feature weights at different time steps and spatial positions. A difference degree calculation unit that uses the Wasserstein distance to measure the state distribution difference in the differences between individual cells, constructs an eigenvalue spectrum analysis model based on the covariance matrix for the group dispersion, and performs dynamic threshold adjustment during the process. According to the operating stage of the battery pack, namely the charging, discharging, and static stages, the decision-making benchmark is automatically corrected, and the output result is a composite difference index that includes a short-term fluctuation index and a long-term degradation trend.

4. The self-maintenance system based on the monitoring of the operating state differences of individual storage batteries according to claim 1, characterized in that, The working process of the adaptive maintenance strategy generator includes: Step S1: Establish a three-dimensional decision space coordinate system, namely the difference degree, maintenance cost, and expected benefit. Step S2: Use Monte Carlo tree search to generate a candidate strategy set. Step S3: Evaluate the Q-value function of each strategy through a value network. Step S4: Introduce an environmental perception compensation mechanism to dynamically adjust the strategy weights according to external factors such as the current environmental temperature and vibration intensity. Step S5: Output the optimal maintenance instruction sequence and the execution timing plan.

5. The self-maintenance system based on the monitoring of the operation state difference of individual storage batteries according to claim 1, wherein It also includes a battery pack life prediction method based on multi-modal data fusion, and the specific steps are as follows: Step 1: Construct a cross-domain feature matrix that includes electrical, thermodynamic, and chemical parameters. Step 2: Train a spatio-temporal graph convolutional network to capture the coupling effect between individual cells. Step 3: Establish an attention-guided long short-term memory model to predict the remaining life; Step 4: Feed the prediction results back to the maintenance strategy generator to dynamically optimize the maintenance cycle parameters; Step 5: Verify the effectiveness of the prediction model through the digital twin system and update the parameters online.

6. The battery pack life prediction method based on multi-modal data fusion according to claim 5, characterized in that The spatio-temporal graph convolutional network in Step 2 includes: Spatial graph convolutional layer: Define the battery connection topology graph as graph-structured data; Temporal convolutional layer: Use dilated convolution to capture long-term temporal dependencies; Dynamic edge weight calculation module: Update the connection weights between nodes according to real-time operation data; Cross-domain feature fusion gate: Integrate the information flow of electrical features and thermodynamic features.

7. The self-maintenance system based on the monitoring of the operating state differences of individual storage batteries according to claim 1, characterized in that It also includes a self-healing communication network, which has a redundant channel dynamic switching mechanism, a hybrid communication protocol based on Lora and CAN bus, a data integrity verification module, and an anti-electromagnetic interference shielding structure.

8. The self-maintenance system based on the monitoring of the operation state differences of individual storage batteries according to claim 1, characterized in that, The pulse energy injection technology is introduced in the maintenance execution stage, which specifically includes a high-frequency resonant equalization circuit, a controllable avalanche energy release device, a nanosecond pulse generator based on GaN devices, and an energy recovery unit.

Images