Consider to implement a more declarative approach for validate_settings. You wrote

Still, I can't think of a better way other than checking each individual key in each individual dictionay/subdictionary, since each key has a different name, each subdict has a different structure and many different data types are used for values.

Of course, but a generic description for this cannot be much more complex than your actual JSON file? Why not have a complementary description file in JSON for those values, for example, like this:

{
  "global_settings": {
    "param1": "integer",
    "param2": "bool",
    "param3": "double|optional"
  },
  "simulation_settings": {
    "param4": "bool",
    "param5": "bool",
    "param6": "bool|optional"
  },
  "output_settings": {
    "file_settings": {
      "param7": "string"
      "param8": "bool"
      "param9": "bool|optional"
    },
  "param10": "bool"
  }
}

Now, validate_settings can use this description and do the validation for data types and non-optional values in a generic fashion. And that makes the actual testing of validate_settings a whole lot simpler, since for getting full test coverage, you can now use a few simple tests which cover each data type once and the optional/non-optional feature.

Of course, you will also have to proofread the description file to make sure it contains no errors, but that is way less effort than writing a test for each original attribute (and proofread also there is no error in those tests).

Consider to implement a more declarative approach for validate_settings. You wrote

Still, I can't think of a better way other than checking each individual key in each individual dictionay/subdictionary, since each key has a different name, each subdict has a different structure and many different data types are used for values.

I can think of one. I guess a generic description for the settings cannot be much more complex than your actual JSON file? Why not have a complementary description file in JSON for those values, for example, like this:

{
  "global_settings": {
    "param1": "integer",
    "param2": "bool",
    "param3": "double|optional"
  },
  "simulation_settings": {
    "param4": "bool",
    "param5": "bool",
    "param6": "bool|optional"
  },
  "output_settings": {
    "file_settings": {
      "param7": "string"
      "param8": "bool"
      "param9": "bool|optional"
    },
  "param10": "bool"
  }
}

Now, validate_settings can use this description and do the validation for data types and non-optional values in a generic fashion. And that makes the actual testing of validate_settings a whole lot simpler, since for getting full test coverage, you can now use a few simple tests which cover each data type once and the optional/non-optional feature.

Of course, you will also have to proofread the description file to make sure it contains no errors, but that is way less effort than writing a test for each original attribute (and proofread also there is no error in those tests).

As an alternative, you could also have a look at JSON schema, which seems to be way more powerful, but I am not an expert on this.