High-sensitivity electronic load

Abstract

The invention discloses a high-sensitivity electronic load, which is characterized in that the current actual output of the electronic load is set as a current calculation value, the expected output of the electronic load is set as a target calculation value, a data network related to the calculation values is established, and in the data network, each calculation value and a corresponding preset adjustment amount form a one-to-one mapping relationship and are combined to form a set; each calculation value forms an incidence relation according to a time sequence, acquiring a current calculation value and a future calculation value at a next time point after being adjusted by a preset adjustment amount, establishing an objective function about the future calculation value, calculating a mean square error between the future calculation value and the objective calculation value, and calculating a mean square error between the future calculation value and the objective calculation value; and performing gradient descent calculation on the mean square error to minimize the mean square error between the future calculation value and the target calculation value, adjusting the discount factor of the preset adjustment amount to obtain a new discount factor, and obtaining a new preset adjustment amount according to the new discount factor.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic loads, and more particularly to a high-sensitivity electronic load. Background Art

[0002] The battery industry has played an increasingly important role in the global energy transformation and development of new energy vehicles in recent years. With the increasing awareness of reducing greenhouse gas emissions and environmental protection around the world, new energy vehicles and their core components, power batteries, have received great attention and rapid development. The industry has a lot of policy support and market demand, and technological innovation has also developed rapidly.

[0003] Among the related industrial products of power batteries, electronic load, also known as electronic load in English, is an electronic component that can accurately detect load voltage and precisely adjust load current by controlling the internal power or conduction amount of transistors. It can simulate the load conditions under actual working conditions during battery debugging and testing. Electronic loads simulate different load conditions to evaluate their discharge characteristics and lifespan, helping to test and evaluate the stability, efficiency and reliability of batteries. The simulation process of electronic loads can help optimize battery design and improve battery performance to ensure that the battery can meet performance requirements in actual use. For battery manufacturers, these are crucial to the development and production of batteries.

[0004] In order to improve the stability and reliability of the battery test process, the electronic load needs to have a suitable expected output, so that the electronic load can quickly adjust the output when the load conditions change and reduce the response time. Therefore, most electronic loads have adaptive control capabilities, which can ensure output stability under different input conditions, avoid overshoot and oscillation, adapt to various uncertainties and interferences, and maintain the reliability of electronic load performance. Summary of the invention

[0005] In order to achieve high sensitivity performance of an electronic load adaptive control process, the present invention provides a high-sensitivity electronic load.

[0006] The technical solution of the present invention is as follows:

[0007] A highly sensitive electronic load,

[0008] Set the current actual output of the electronic load as the current calculation value and the expected output of the electronic load as the target calculation value.

[0009] A data network related to the calculated values ​​is established. In the data network, each calculated value and the corresponding preset adjustment amount form a one-to-one mapping relationship and are combined to form a set. Each calculated value forms an association relationship in time sequence.

[0010] Obtain the current calculated value and the future calculated value at the next time point after adjustment by the preset adjustment amount, establish an objective function for the future calculated value, and include a discount factor for the preset adjustment amount in the objective function.

[0011] Calculate the mean square error between the future calculated value and the target calculated value, and perform gradient descent calculation on the mean square error to minimize the mean square error between the future calculated value and the target calculated value, adjust the discount factor of the preset adjustment amount, obtain a new discount factor, and obtain a new preset adjustment amount according to the new discount factor.

[0012] In the above-mentioned high-sensitivity electronic load, the content of the preset adjustment amount includes an adjustment value for adjusting the performance parameter of the electronic load and a discount factor for calculating the updated preset adjustment amount.

[0013] A single training process of the self-adjustment calculation of the electronic load of the above-mentioned high-sensitivity electronic load includes:

[0014] Acquire the current characteristic state of the electronic load, determine the current preset adjustment amount of the electronic load according to the electronic load self-adjustment calculation method, and perform adjustment on the current characteristic state of the electronic load.

[0015] Obtain the updated characteristic state of the electronic load, calculate the difference between the actual output of the electronic load update and the expected output,

[0016] Determine whether the electronic load update characteristic state meets the training setting conditions. If so, stop training; if not, continue training.

[0017] Furthermore, the training setting conditions include

[0018] (1) The number of training rounds of the electronic load's self-adjustment calculation reaches a preset fixed number of training rounds;

[0019] (2) A characteristic parameter of the updated characteristic state of the electronic load reaches a preset value;

[0020] (3) The training time calculated by the self-adjustment of the electronic load reaches the preset training time;

[0021] (4) The difference between the actual output updated by the electronic load and the expected output reaches a preset threshold;

[0022] (5) External control signal controls the electronic load training to stop;

[0023] (6) The working performance of the electronic load reaches the preset value.

[0024] For the above-mentioned high-sensitivity electronic load, the objective function is y = γQ t+1 ,

[0025] Among them, y is the objective function,

[0026] γ is the discount factor,

[0027] Q t+1 Calculates a value for the future.

[0028] Furthermore, the formula for calculating the mean square error between the future calculated value and the target calculated value is:

[0029] U=(yQ) 2 =(γQ t+1 -Q) 2 ,

[0030] Among them, U is the mean square error between the future calculated value and the target calculated value,

[0031] Q is the target calculation value.

[0032] Furthermore, the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor is calculated. γ U, ▽ γ The calculation formula of U is

[0033] ▽ γ U=2·Q t+1 ·(γQ t+1 -Q),

[0034] Among them, γ U is the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U,

[0035] Q t+1 To calculate the value for the future,

[0036] Q is the target calculation value.

[0037] Furthermore, the formula for calculating the mean square error between the future calculated value and the target calculated value is:

[0038]

[0039] Among them, U is the mean square error between the future calculated value and the target calculated value,

[0040] Q is the target calculation value,

[0041] N is the total number of all future calculated values ​​within a certain period of time.

[0042] Furthermore, the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor is calculated. γ U, ▽ γ The calculation formula of U is

[0043]

[0044] Among them, γ U is the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U,

[0045] Q t+1 To calculate the value for the future,

[0046] Q is the target calculation value,

[0047] N is the total number of all future calculated values ​​within a certain period of time.

[0048] The above-mentioned high-sensitivity electronic load replaces the original preset adjustment amount with the updated preset adjustment amount and saves it in the corresponding set, or sets the priority of the preset adjustment amount according to the generation time. For the preset adjustment amount corresponding to the same current calculated value, the later the generation time is, the higher the adoption priority is.

[0049] The present invention according to the above scheme has the beneficial effect that:

[0050] 1. The adjustment amount for the current characteristic state of the electronic load of the present invention is a preset adjustment amount, which is saved in conjunction with the current calculated value of the electronic load. The electronic load can quickly obtain the preset adjustment amount by simply calling out a set of saved calculated values ​​that are the same as the current calculated value, and directly calling the value to adjust the electronic load, thereby improving the response sensitivity of the electronic load.

[0051] 2. The preset adjustment amount is a post-variable amount. As time goes by, the system gradually stabilizes, and the preset adjustment amount obtained in this way also gradually stabilizes, making the actual output of the electronic load approach the expected output, ensuring the rapid calling of the preset adjustment amount while maintaining long-term stability. DETAILED DESCRIPTION

[0052] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] A highly sensitive electronic load,

[0054] Set the current actual output of the electronic load as the current calculation value and the expected output of the electronic load as the target calculation value.

[0055] A data network related to the calculated values ​​is established. In the data network, each calculated value and the corresponding preset adjustment amount form a one-to-one mapping relationship and are combined to form a set. Each calculated value forms an association relationship in time sequence.

[0056] Obtain the current calculated value and the future calculated value at the next time point after adjustment by the preset adjustment amount, establish an objective function for the future calculated value, and include a discount factor for the preset adjustment amount in the objective function.

[0057] Calculate the mean square error between the future calculated value and the target calculated value, and perform gradient descent calculation on the mean square error to minimize the mean square error between the future calculated value and the target calculated value, adjust the discount factor of the preset adjustment amount, obtain a new discount factor, and obtain a new preset adjustment amount according to the new discount factor.

[0058] The content of the preset adjustment amount includes an adjustment value for adjusting the performance parameter of the electronic load and a discount factor for calculating an updated preset adjustment amount.

[0059] The adjustment amount of the current characteristic state of the electronic load of the present invention is a preset adjustment amount (the preset adjustment amount is the adjustment value of the current characteristic state of the electronic load, and the parameter value of the updated characteristic state of the electronic load is equal to the sum of the parameter value of the current characteristic state and the preset adjustment amount), and the preset adjustment amount is a relatively fixed value. Although the preset adjustment amount is actually changing at every moment through the subsequent mean square error calculation, gradient calculation, and discount factor calculation, for the adjustment of the electronic load corresponding to the current calculated value, this is a fixed value, which is saved in conjunction with the current calculated value of the electronic load. At that time, as long as the set of saved calculated values ​​that are the same as the current calculated value is called out, the preset adjustment amount can be quickly obtained, and the value can be directly called to adjust the electronic load, thereby improving the responsiveness of the electronic load.

[0060] In the initial state, the electronic load is preset with a set of calculated values ​​and initial preset adjustment values. There are many sets, which can be sourced from previous electronic load test data, manual presets, or data obtained from simulation tests. In short, these calculated values ​​and corresponding preset adjustment values ​​should all be theoretically existing values, so that the electronic load can run stably in the initial stage. In addition, the number of sets stored in the electronic load needs to be determined based on accuracy. Within the required range of a certain test, as many of these calculated values ​​as possible should be included according to the test accuracy for easy call. In practical applications, the electronic load obtains its actual output value through the monitoring system. This value is the current calculated value. The electronic load searches for the calculated value that is the same as the current calculated value in its own system, calls out the content of the set to which it belongs, obtains the preset adjustment amount in the memory, and adjusts the working parameters of the electronic load according to the adjustment value in the preset adjustment amount. Then, as time goes by, a future calculated value is obtained. The future calculated value and the target calculated value are subjected to mean square error, gradient descent, mean square error minimization and other calculations to obtain a new discount factor. A new preset adjustment amount is calculated according to the new discount factor, and then the preset adjustment amount is saved. The future calculated value becomes the current calculated value, and the above calculation is repeated.

[0061] The objective function is calculated as follows.

[0062] The objective function is y = γQ t+1 ,

[0063] Among them, y is the objective function,

[0064] γ is the discount factor,

[0065] Q t+1 Calculates a value for the future.

[0066] The formula for calculating the mean square error between the future calculated value and the target calculated value is:

[0067] U=(yQ) 2 =(γQ t+1 -Q) 2 ,

[0068] Among them, U is the mean square error between the future calculated value and the target calculated value,

[0069] Q is the target calculation value.

[0070] Calculate the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U, ▽ γ The calculation formula of U is

[0071] ▽ γ U=2·Q t+1·(γQ t+1 -Q),

[0072] Among them, γ U is the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U,

[0073] Q t+1 To calculate the value for the future,

[0074] Q is the target calculation value.

[0075] The formula for calculating the mean square error between the future calculated value and the target calculated value is:

[0076]

[0077] Among them, U is the mean square error between the future calculated value and the target calculated value,

[0078] Q is the target calculation value,

[0079] N is the total number of all future calculated values ​​within a certain period of time.

[0080] Calculate the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U, ▽ γ The calculation formula of U is

[0081]

[0082] Among them, γ U is the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U,

[0083] Q t+1 To calculate the value for the future,

[0084] Q is the target calculation value,

[0085] N is the total number of all future calculated values ​​within a certain period of time.

[0086] The above are two calculation methods for adjusting the discount factor. The first calculation method for adjusting the discount factor is a single-value discount factor adjustment algorithm for the current calculated value. The second calculation method for adjusting the discount factor is a multiple-value discount factor adjustment algorithm for all current calculated values ​​within a certain time period. The difference lies in the calculation of the mean square error between the future calculated value and the target calculated value. The former only takes a single value for calculation, while the latter takes the average of the mean square errors of multiple values.

[0087] The first calculation method for adjusting the discount factor only considers a single value each time, so that the electronic load can quickly respond to the current load changes, which is conducive to quickly adjusting the preset adjustment amount under rapidly changing load conditions to ensure the timeliness and accuracy of the output. In addition, each adjustment is based on the current error, so the local optimal discount factor can be found, allowing the electronic load to quickly converge to a more ideal control state in a short time.

[0088] The second calculation method for adjusting the discount factor can better capture the overall trend and avoid misadjustment due to local fluctuations by considering the mean square error of multiple values, making it easier to find the global optimal discount factor and improve the overall control performance. It can gradually adjust the discount factor and is more suitable for long-term optimization, making the electronic load more adaptable to long-term changing load conditions.

[0089] Since the second calculation method for adjusting the discount factor can set the time period arbitrarily, a shorter time period within the allowable range can be set to appropriately reduce the sensitivity, reduce the functional requirements for computing and control chips, reduce the cost of electronic loads, and avoid excessive and complicated calculations.

[0090] After obtaining the updated preset adjustment amount, the updated preset adjustment amount replaces the original preset adjustment amount and is saved in the corresponding set, or the priority of the preset adjustment amount is set according to the generation time. For the preset adjustment amount corresponding to the same current calculation value, the later the generation time, the higher the adoption priority.

[0091] After obtaining the updated discount factor, the update calculation of the preset adjustment amount is: the ratio of the original preset adjustment amount to the updated preset adjustment amount is equal to the ratio of the original discount factor to the updated discount factor, and the two are in a positive proportional relationship.

[0092] After calculating the new preset adjustment amount, there are two solutions for processing this preset adjustment amount: one is to replace the original value. In this way, each updated preset adjustment amount takes effect immediately. The update and management process is simple. It only needs to replace the corresponding value in the set. It can quickly reflect the current load changes. The program does not require additional data structures or complex logic, which saves computing resources and storage space and reduces the difficulty of development and maintenance. The other is to set up a system for saving historical data and adopting it with priority. This method retains all generated preset adjustments and their generation time. Historical data can be traced back when needed, which is conducive to troubleshooting and data analysis. In addition, when the electronic load is subsequently improved, the adjustment data and adjustment effects at different time points are analyzed using historical data, which can provide analysis content for the adjustment of the electronic load.

[0093] In order to stabilize the preset adjustment amount of the electronic load during the actual test process, the electronic load can be trained for self-adjustment calculation in advance under the same state and standard input. The single training process of the electronic load self-adjustment calculation includes:

[0094] Acquire the current characteristic state of the electronic load, determine the current preset adjustment amount of the electronic load according to the electronic load self-adjustment calculation method, and perform adjustment on the current characteristic state of the electronic load.

[0095] Obtain the updated characteristic state of the electronic load, calculate the difference between the actual output of the electronic load update and the expected output,

[0096] Determine whether the electronic load update characteristic state meets the training setting conditions. If so, stop training; if not, continue training.

[0097] The training setting conditions include

[0098] (1) The number of training rounds of the electronic load's self-adjustment calculation reaches a preset fixed number of training rounds;

[0099] (2) A characteristic parameter of the updated characteristic state of the electronic load reaches a preset value;

[0100] (3) The training time calculated by the self-adjustment of the electronic load reaches the preset training time;

[0101] (4) The difference between the actual output updated by the electronic load and the expected output reaches a preset threshold;

[0102] (5) External control signal controls the electronic load training to stop;

[0103] (6) The working performance of the electronic load reaches the preset value.

[0104] The self-adjustment algorithm of the electronic load is actually in continuous training. Each batch of training is divided based on the training setting conditions. When the training process of the self-adjustment algorithm triggers a certain training setting condition, the current batch of self-adjustment algorithms stops and a new round of self-adjustment algorithms is restarted according to the settings.

[0105] The explanation of the training setting conditions is as follows.

[0106] (1) The number of training rounds of the self-adjustment calculation of the electronic load reaches a preset fixed number of training rounds: The electronic load has a fixed number of training rounds set in the system in advance, such as 1000 rounds of training. When the 1000th round of training of the self-adjustment calculation of the electronic load is completed, the training process ends.

[0107] (2) A characteristic parameter of the updated characteristic state of the electronic load reaches a preset value: The electronic load pre-sets a performance standard in the system. When the performance of the model reaches or exceeds this standard, the training process ends. For example, if the electronic load pre-sets 24W as the preset power value in the system, when the actual output power of the electronic load reaches 24W or higher, the training process of this batch ends; if the electronic load pre-sets 240A as the preset current value in the system, when the actual output current of the electronic load reaches 240A or higher, the training process of this batch ends.

[0108] (3) The training time of the self-adjustment calculation of the electronic load reaches the preset training time: The electronic load sets a fixed time limit in the system in advance. When the training time reaches this limit, the training process ends. For example, if we set the training time to 24 hours, when the training time of the self-adjustment calculation of the electronic load reaches 24 hours, the training process of this batch ends.

[0109] (4) The difference between the actual output of the electronic load and the expected output reaches a preset threshold: The electronic load sets a preset threshold in the system as the preset threshold for updating the difference between the actual output and the expected output. When the difference between the actual output of the electronic load and the expected output reaches or falls below the preset threshold, the training process ends. For example, if the difference between the actual output of the electronic load and the expected output in the system is less than a certain threshold value - 0.1, when the average difference between the actual output of the electronic load and the expected output reaches or falls below 0.1, or when the difference between the actual output of the electronic load and the expected output is equal to or less than 0.1, the training process ends.

[0110] (5) External control signal controls the electronic load training to stop: This machine is set up or connected to manual input hardware so that the machine receives a stop control signal, or the control terminal transmits a stop control signal to the machine. After the machine receives the signal, the training process ends.

[0111] (6) The working performance of the electronic load reaches the preset value: In addition to the power, current, and voltage that play the main role in the work of the electronic load itself, there are also performance parameters that indicate its own working environment or its own working state, such as ambient temperature, hardware temperature, its own working voltage, its own working current, its own working power, its own working current fluctuation, etc. The electronic load pre-sets a certain threshold in the system as the limiting value of the performance parameter. When the working state of the electronic load reaches or exceeds the threshold, the training process of this batch ends. For example, if the threshold value of the working temperature of the electronic load is set to 80 degrees Celsius, when the working temperature of the electronic load (the temperature of a certain hardware of the electronic load) reaches or exceeds 80 degrees Celsius, the training process ends.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A highly sensitive electronic load, characterized in that: Set the current actual output of the electronic load as the current calculation value and the expected output of the electronic load as the target calculation value. A data network related to the calculated values ​​is established. In the data network, each calculated value and the corresponding preset adjustment amount form a one-to-one mapping relationship and are combined to form a set. Each calculated value forms an association relationship in time sequence. The content of the preset adjustment amount includes an adjustment value for adjusting the performance parameter of the electronic load and a discount factor for calculating the updated preset adjustment amount. Obtain the current calculated value and the future calculated value at the next time point after adjustment by the preset adjustment amount, establish an objective function for the future calculated value, and include a discount factor for the preset adjustment amount in the objective function. Calculate the mean square error between the future calculated value and the target calculated value, and perform gradient descent calculation on the mean square error to minimize the mean square error between the future calculated value and the target calculated value, adjust the discount factor of the preset adjustment amount, obtain a new discount factor, and obtain a new preset adjustment amount according to the new discount factor.

2. A high-sensitivity electronic load according to claim 1, characterized in that: A single training process for the electronic load's self-tuning calculations includes Acquire the current characteristic state of the electronic load, determine the current preset adjustment amount of the electronic load according to the electronic load self-adjustment calculation method, and perform adjustment on the current characteristic state of the electronic load. Obtain the updated characteristic state of the electronic load, calculate the difference between the actual output of the electronic load update and the expected output, Determine whether the electronic load update characteristic state meets the training setting conditions. If so, stop training; if not, continue training.

3. A high-sensitivity electronic load according to claim 2, characterized in that: The training setting conditions include (1) The number of training rounds of the electronic load's self-adjustment calculation reaches a preset fixed number of training rounds; (2) A characteristic parameter of the updated characteristic state of the electronic load reaches a preset value; (3) The training time calculated by the self-adjustment of the electronic load reaches the preset training time; (4) The difference between the actual output updated by the electronic load and the expected output reaches a preset threshold; (5) External control signal controls the electronic load training to stop; (6) The working performance of the electronic load reaches the preset value.

4. A high-sensitivity electronic load according to claim 1, characterized in that: The objective function is y = γQ t+1 , Among them, y is the objective function, γ is the discount factor, Q t+1 Calculates a value for the future.

5. A high-sensitivity electronic load according to claim 4, characterized in that: The formula for calculating the mean square error between the future calculated value and the target calculated value is: U=(y-Q) 2 =(γQ t+1 -Q) 2 , Among them, U is the mean square error between the future calculated value and the target calculated value, Q is the target calculation value.

6. A high-sensitivity electronic load according to claim 6, characterized in that: Calculate the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U, ▽ γ The calculation formula of U is ▽ γ U=2·Q t+1 ·(γQ t+1 -Q), Among them, γ U is the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U, Q t+1 To calculate the value for the future, Q is the target calculated value.

7. A high-sensitivity electronic load according to claim 4, characterized in that: The formula for calculating the mean square error between the future calculated value and the target calculated value is: Among them, U is the mean square error between the future calculated value and the target calculated value, Q is the target calculation value, N is the total number of all future calculated values ​​within a certain period of time.

8. A high-sensitivity electronic load according to claim 7, characterized in that: Calculate the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U, ▽ γ The calculation formula of U is Among them, γ U is the gradient of the mean square error between the future calculated value and the target calculated value with respect to the discount factor ▽ γ U, Q t+1 To calculate the value for the future, Q is the target calculation value, N is the total number of all future calculated values ​​within a certain period of time.

9. A high-sensitivity electronic load according to claim 1, characterized in that: The content of the preset adjustment amount includes an adjustment value for adjusting the performance parameter of the electronic load and a discount factor for calculating an updated preset adjustment amount.

10. A high-sensitivity electronic load according to claim 1, characterized in that: The updated preset adjustment amount replaces the original preset adjustment amount and is saved in the corresponding set, or the priority of the preset adjustment amount is set according to the generation time. For the preset adjustment amount corresponding to the same current calculated value, the later the generation time is, the higher the adoption priority is.

Images